Changelog

What’s New (last 14 days)

Narrative summary unavailable. See Full history below.

Full history

2026-05

Features

wire llms_txt + llms_full_txt generators (WG-49) (#1540)
WG-48 wire changelog_generator (#1528) (#1539)
wire architecture_doc_generator (WG-47) (#1537)
add generators.enabled[] config knob (WG-45) (#1536)
Hermes research — operational documentation + label ownership (#1521)
Hermes research — Telegram bot handlers (#1520)
research bisync wrapper + comment helper (#1417) (#1505)
restore codex-wrapper.sh as opt-in adversarial backend (#1458) (#1501)
extract fix.md §3.8 recovery search to fix-recovery-search.sh (#1499)
add scripts/research-suggest.js (child E of #1380) (#1476)
document Hermes research Drive folder setup (#1472)
wiki-generator: exempt allowlisted public hosts from secret scan() (#1469)
block raw gh issue create to enforce issue-create.js pipeline (#1471)
fast-path on-session-start.sh for non-interactive Claude Code invocations (#1466)
[wiki-generator] WG-21 — end-to-end integration test (#1467)
[wiki-generator] WG-29 — Cost monitor (#1463)
Hermes research — polling worker (FIFO + lockfile) (#1455)
[wiki-generator] WG-32 — CLI / manual invocation (#1456)
[wiki-generator] WG-20 — Orchestrator + DI hooks (#1451)
Hermes research issue body validator + GH Action (#1452)
[wiki-generator] WG-16 follow-up: harden architecture doc generator (QA gaps) (#1453)
Hermes research — headless invocation wrapper (#1449)
tighten isPlainObject() in secret_sanitizer opts guard (#1450)
paginate wiki-generator issue comments beyond 100 (#1445)
[wiki-generator] WG-18 — Hallucination guard con golden snapshots (#1444)
WG-14 module doc generator (Sonnet 4.6 medium) (#1430)
centralize claude CLI spawn via runClaude helper (#1306) (#1425)
dynamic question budget by complexity tier in /jkz:start (#1440)
WG-19 publisher modules (#1256) (#1437)
harden issue_extractor CLI paths (—cursor / —output) against traversal (#1436)
[wiki-generator] WG-34 — Reference & stats pages generator (#1428)
WG-17 issue-driven content pipeline (#1429)
WG-15 workflow doc generator (#1269) (#1423)
[wiki-generator] WG-35 — Changelog generator (What’s New + Full History) (#1422)
WG-16 architecture doc generator (#1270) (#1424)
[wiki-generator] WG-05 — implementation + PII + issue-log sanitizers + composer (#1400)
WG-10 issue extractor + docs-worthy label (#1389)
[wiki-generator] WG-04 secret sanitizer + provider regexes + adversarial suite (#1388)
WG-25 Pagefind validation via Hermes lite-runner (#1373)
[wiki-generator] WG-13 — Output formatters (sidebar + llms.txt + llms-full.txt) (#1365)
adversarial-bypass defenses for the alignment validator (#1360)
persist background tasks (crons + note) in session snapshot (#1346)
validate paths.wikiOutput against directory traversal (#1345)
Agent View integration — discoverable subagents + worktree yield (#1344)
WG-04 sanitizer helpers (entropy + log + index) [#1294] (#1341)
URI-encode firstFile per segment in editUrl (#1340)
WG-12 — API reference generator (mechanical, zero-LLM) (#1330)
[wiki-generator] WG-04 — Path blocklist (foundation) (#1332)
Ollama provider fallback for Gemini-based roles (#1326)
Consolidate issue creation into shared script primitive with universal adversarial coverage (#1318)
[wiki-generator] WG-31 — State file management & pruning (#1316)
WG-08 AST + JSDoc/TSDoc extractor (#1304)
[wiki-generator] WG-07 — Core utilities (LLM + GitHub + Telegram + pricing wrappers) (#1298)
WG-11 diff & change detection + module hasher (#1300)
[wiki-generator] WG-09 — Lightweight extractors (README + changelog + config) (#1299)
[wiki-generator] WG-30 — Configuración centralizada de la skill (#1289)
issue alignment validator (#1239) (#1241)
expand documentation coverage beyond commands/scripts/agents/skills (#1240)
sentinel meta-audit layer — detect sycophantic Judge patterns (#1200) (#1229)
require L1/L2 evidence before reinforcing PASS patterns (#1231)
mutation: triage surviving mutants in scripts/validators/rules/dry-check.js (#1217)
add industry ratio benchmarks reference table (#1216)
improve test coverage gaps in extract-findings (surviving mutants from Batch A) (#1213)
feat(mutation): improve test coverage gaps in slo-check (surviving mutants from Batch A) (#1210)
improve normalize-judge-verdict test coverage to 83.19% (#1207)
expand Stryker to 4 critical modules (#1185)
Batch D+E — analysis & utility modules (cluster-findings, error-grouping, rubric-score, parse-gemini-stream, dependency-resolve, saturation-check, rollback-decision) (#1201)
Batch C — validators suite (#1192)
Batch A — decision-logic core (vote-resolver, loop-guard, slo-check, plan-digest, extract-findings) (#1191)
Batch B — context-protection layer (truncate-output, compress-context, format-patterns, compress-git-output) (#1181)
add Division of Labor to Judge, Sentinel, Inspector (#1180)
/jkz:cost —attribute — capture orchestrator chat tokens against a PR (#1174)
run-research — add HN, GitHub Issues, Lobsters, Dev.to sources (#1179)
complete tracing integration – tokens, cost, duration, Doctor nesting (#1130) (#1172)
Prometheus /metrics endpoint (#1129) (#1170)
per-token scope-based authorization (read/write/admin) (#1169)
post cost-report PR comment on every merge (#1166)
add HTTP transport with bearer auth to jkz MCP server (#1161)
add source field to skill frontmatter (#1160)
emit memory/skill events to Telegram in real-time (topic “logs”) (#1159)
auto-clean gitignored junk + install post-merge hook (#1157)
add cost-report PR comment hook to /jkz:quick and /jkz:qa (#1158)
post pipeline cost as PR comment for Hermes Telegram relay (#1156)
skill update-readme — 3 parallel agents (#1118) (#1152)
skill repo-to-markdown — convert GitHub repos to single markdown for single-pass analysis (#1150)
skill update-claudemd — bidirectional CLAUDE.md reconciliation with 7-phase workflow (#1134)
skill convert-to-markdown — webpage to markdown via EzyCopy (#1132)
skill make-changelog — parallel Haiku subagents per version range (#1133)
atomic writes for shared global resources (skills + pipeline state) (#1111)
automatic checkpoint snapshots before each pipeline phase transition (#1112)
add ac_item_count signal and remove body truncation (#1113)
add native GitHub blocked-by relationships via GraphQL (#1110)
Periodic exportable memory backups (#1101) (#1109)
implement audit log for memory operations (#1098) (#1105)
CR history persistence — backfill, post-merge ingestion, and Hermes catchup cron (#1107)
add namespace support to memory tool (#1100) (#1108)
builder preflight sub-agent validates acceptance criteria before Judge (#1095) (#1104)
add version history for memory entries (#1099) (#1106)
inject failure taxonomy block into Doctor prompt (#1096)

Fixes

wiki-generator: secret_sanitizer high_entropy fallback flags allowlisted-host URL paths (blocks changelog publish / WG-48) (#1550)
re-read PR_DIFF_CONTEXT in qa.md §3 defensive block (#1546) (#1548)
bind GraphQL Int! variable via gh -F in check-parent-completion.js (#1547)
gh-sub-issue / gh-blocked-by GraphQL Int! variable binding (#1545)
automated git sync of main on Hermes VPS (#1515) (#1544)
hermes-report-wrapper survives foreign-UID-owned report dir (#1543)
omit editUrl/deprecated_since when null (WG-44) (#1535)
WG-40 infer primitive return types (boolean/number/string) and drop ’: unknown’ in docs (#1519)
wiki-generator wraps async return types in Promise<> (#1512) (#1517)
wiki-generator: editUrl: null breaks Starlight schema (cross-repo publish) (#1518)
WG-37 allowlist project public hosts in secret sanitizer (#1509) (#1510)
emit publish paths under wiki/ prefix (WG-36) (#1508)
generate /etc/cron.d/jkz from cron-registry.json (#1492) (#1498)
wire jkz-docs build to docs.j0kz.dev publish target (#1497)
infer returnType for exports without explicit annotation (#1495)
add changelog + issueCategories to config schema (#1496)
wiki-generator signature blocks emit text language tag (MD040) (#1488)
expand glob ignore list to include .next/.cache/coverage/.git (#1485)
wiki-generator editUrl wiring leaks private repo and trips sanitizer (blocks #1276) (#1483)
show publish section in config debug output (#1481)
route publisher to publish.repoUrl, not site.repoUrl (#1480)
exclude node_modules/dist/build from modules glob (#1478)
accept URL-safe base64 cursor characters (#1477)
authenticate LLM client via OAuth, not API key (#1473)
add missing run subcommand to wiki-generator cron (#1470)
dedupe complexity:* labels in mergeLabels (#1464)
[wiki-generator] phaseSanitize reads commit.message but extractor emits commit.subject -> secret scan inert (#1462)
[wiki-generator] phaseGuard re-screens trusted frontmatter editUrl -> every generated page fails guard (#1461)
[mcp] resolve npm audit vulnerabilities (#1315) (#1454)
expand heredoc placeholders so PR diff reaches Lens/Sentinel (#1433)
prevent jkz-update tests from overwriting tracked files (T11/T12) (#1411)
decode JSON envelopes in extract-comment-block (#1374) (#1409)
detach watchdog stdout to prevent command-substitution hang (#1407)
worktree post-merge cleanup leaks (3 fixes) (#1403)
guard JKZ_GEMINI_MODEL emission with V_GEMINI_API_ENDPOINT check (#1395)
replace STATE.json reverse-pointer with GitHub native sub-issues GraphQL (#1396)
watchdog for wrapper post-completion hang (#1387) (#1394)
build-pr-diff.js regex corrupts patch content (][ → ,), poisons every PR review (#1397)
resolve architect deliberation by issue number (#1375) (#1383)
replace echo|grep with bash pattern match (fixes two false-negatives) (#1382)
bump fast-uri to 3.1.2 and ip-address to 10.2.0 (#1369)
pipe PR comment cache via stdin to avoid E2BIG (#1337) (#1342)
restore brief metadata + pipeline persistence dropped during #1302 refactor (#1331)
use version subcommand and ship sqlite3 in container (#1313)
auto-register native GitHub blocked-by + sub-issue relationships at issue creation (#1286) (#1305)
add pr-merge-label-sync workflow (#1292) (#1297)
make cost-report relay first-writer-wins (#1291)
bake complexity:* into gh issue create via shared helper (#1234)
doc-sync: implement issue creation when discrepancies detected (#1235)
post comment without token data + orchestrator attribution in quick (#1212)
migrate agent emails to j0kz.dev domain
persist webhook [SILENT] patch and jobs.json ownership across rebuilds (#1162)
/jkz:quick lock TTL + PR-number cascade for agent worktrees (#1153)
restore test suite — fix 24 pre-existing failures (#1147) (#1151)
task-discovery skips complexity label on non-high confidence (#1149)
Telegram bot: race condition in checkpoint approval (callback + reply) (#1141)
telegram-bot sendMessage truncation strips parse_mode and logs tgApi failures (#1145)
Telegram bot: pendingIdCounter resets on restart, IDs collide with previous-session buttons (#1144)
correct phase ordering + restore computeCost fallback (#1139)
route adversarial detection through resolve-wrapper.sh (#1128)
use exit 2 for runtime errors in orchestrate and jkz-update (#1123)

Refactors

extract qa.md §1 context-gathering into qa-gather-context.sh (Pattern B, #1358) (#1538)
extract §1.6 failure classifier to Pattern B library (#1357) (#1522)
wire per-folder bisync after research-comment (#1507)
extract §1.3 flaky classifier to scripts/lib/fix-classify-flaky.sh (#1502)
extract Auditor dispatch to scripts/lib/plan-invoke-auditor.sh (#1352) (#1500)
mode-aware postcondition checks (fixes #1406) (#1489)
extract fix.md §7.3 journal renderer to scripts/lib/fix-journal-update.sh (#1446)
triage 2026-05-11 entropy scan (#1339) (#1438)
release worktree lock in /jkz:quit when pipeline approved (#1432) (#1434)
migrate non-skill callers to issue-create.js (#1426)
shift-left engineering disciplines (named exceptions, shadow paths, retro check) (#1408)
docs(index): restructure INDEX.md by Diátaxis quadrant + tracked gaps (#1404)
extract section 1.8 memory queries into fix-query-memory.sh (Pattern B pioneer) (#1355) (#1385)
extract Architect prompt §3 assembly to scripts/lib/plan-build-architect-prompt.sh (#1384)
extract §3 Lens + Sentinel dispatch to scripts/lib/ helpers (#1381)
port tier-guard + triage detection from legacy issue.md (#1324)

Other

use authenticated gh bootstrap for private-repo install (#1563)
remove orphan wiki-sync.yml workflow (#1503)
wrap nested fence in module_doc.md prompt (#1442) (#1475)
broaden AC-7 isolation discovery to .mjs/.cjs (#1474)
cover server-side cursor-error fallback in issue_extractor (#1468)
update wiki-generator deps (ts-morph v28, octokit v5, anthropic-sdk 0.95) (#1465)
[wiki-generator] WG-28 — CI quality gates suite + llms.txt validator (#1448)
[wiki-generator] WG-22 — Hermes task config (#1457)
cover PERSIST_ERR write-failure branch (T9) (#1447)
add deepseek-v4-pro and kimi-k2.6 cloud pricing (#1443)
migrate validator default from 3.1-pro-preview to 3.5-flash (#1439)
bump the minor-and-patch group with 4 updates (#1402)
bump the minor-and-patch group across 1 directory with 2 updates (#1435)
translate ADR-008 headers to English (#1379) (#1410)
adopt Ollama Cloud as documented default for Curator + Research-Reviewer (#1363) (#1378)
Diátaxis classification of 172 markdown files (#1372)
add Diátaxis index audit and refactor plan (#1368)
add backup recovery walkthrough (#1296) (#1366)
cover pageNameForDir for multi-segment dirRel (#1335) (#1359)
stop hardcoding JKZ_CODEX_MODEL in managed env block (#1362)
script-extraction plan for jkz command refactor (#1361)
cover strict ISO-8601 now rejection paths (#1348)
extend provider abstraction plan with Task 6 (doc-reviewer issue) (#1347)
provider abstraction design + 5-issue implementation plan (#1343)
align mocks with create-issue-with-complexity helper (#1338)
bump the minor-and-patch group with 2 updates (#1328)
bump @hono/node-server from 1.19.14 to 2.0.2 in /mcp (#1329)
bump @types/node in /mcp in the minor-and-patch group (#1327)
document DEPLOY_WEBHOOK_URL secret in runbook (#1325)
close test, migration, and dead-code gaps from #1302 refactor (#1323)
backfill native GitHub relationships for wiki-generator issues (#1322)
Hermes VPS backup with restic + GDrive (14d/4w/3m) (#1296) (#1310)
strip Codex as default runtime backend (#1308)
[wiki-generator] WG-27 — GitHub Wiki sync action (#1301)
WG-33 skill versioning + CHANGELOG (#1246) (#1290)
WG-03 Telegram topics + WIKI_BOT_TOKEN runbook (#1244) (#1288)
WG-02 skill scaffolding (#1243) (#1285)
extend wiki-generator v3.1 with WG-34/WG-35 and 05:00 schedule (#1284)
doc-sync: sanitize MCP scanner args to avoid surfacing potential secrets (#1281)
add wiki-generator pipeline plan v3.1 (#1275)
trivial follow-ups from #1223 QA (L1, L2) (#1280)
add Conway’s Law mapping to README and explainers (#1238)
remove quantconnect server (#1237)
mutant-driven tests — raise score from 19.72% to 95.77% (#1232)
triage surviving mutants in dependency-resolve.js (88% -> 92%) (#1236)
rubric-score.js triage — Batch D+E (56→96%) (#1230)
improve error-grouping.js mutation coverage (#1224)
triage surviving mutants in capabilities.js — achieve ≥80% score (closes #1198) (#1219)
mutant-driven tests — raise score from 37.67% to 95.21% (#1218)
triage surviving mutants in test-coverage.js (>=80% gate) (#1220)
improve validators/run.js mutation coverage to 81.37% (Batch C) (#1215)
add tests to kill plan-digest.js surviving mutants (#1214)
improve loop-guard.js mutation coverage (#1211)
improve vote-resolver mutation coverage (issue #1186) (#1209)
improve step-gate.js coverage to 87.63% (#1208)
improve mutation coverage for parse-body-deps.js (#1206)
add tests killing 19 surviving mutants (mutation score 56.82% -> 100%) (#1171)
bump the minor-and-patch group across 1 directory with 2 updates (#1148)
verify j0kz.dev attribution
add ADR-008 and ignore Python bytecode (#1168)
harden migration safety scanner (Judge LOW findings, #1164) (#1165)
add SQLite migration safety gate to CI (#1124) (#1163)
document source-availability reality (#1121) (#1154)
spike mutation testing on atomic-write.js (Stryker) (#1143)
refactor: condense CLAUDE.md — remove redundant one-liners, establish 130-line ceiling (#1131)

2026-04

Features

rollback support for jkz update (—version, —list, —previous) (#1087)
add chore as first-class issue type (#1072) (#1086)
per-project version pinning via .jkz-version (#1066) (#1085)
CI gate: validate tags before publishing GitHub Release (#1084)
publish tarball + SHA256 to GH Releases, verify on update (#1083)
adopt tagged releases for jkz update (#1077)
CR thread resolution gate + atomic resolve helper (#1075)
add per-type issue templates and load them in /jkz:issue (#1080)
Adopt Keep a Changelog + explicit semver policy for jkz (#1079)
extract shared git-root wrapper to dedupe settings.json prefixes (#1073)
harden per-project sub-containers for non-root UID (#1060)
hermes(docker): run container as non-root user (#1051)
global workspace mode — multi-project desktop view (#993) (#1045)
add typed pushEvent API for one-shot frames (#1048)
cloud relay client refactor + relay contract docs (#1032)
installer for jkz-multi-agent skill and jkz-dev profile (#1034)
on-demand /jkz:cost skill with CR overhead tracking (#1031)
sync token_usage to Hermes + phase breakdown (#1030)
add isRefetching state to useGraph for polling UX (#1014) (#1023)
add rate limiting to /api/graph endpoint (#1022)
cloud relay hook — outbound WSS to ContentCurator (#1020)
pipeline graph view — D3 force-directed graph with dependency visualization (#1008)
e2e Opus cost tracking via SubagentStop hook + pipeline skill fallback (#1012)
monitoring panel — rate limits + active sessions (#1006)
knowledge base viewer — deliberations, patterns, memory (#992) (#1007)
Hermes post-merge cost report to Telegram topic (#1005)
Chat isolation activation: inject worktree-gate into pipeline commands (#1002)
native worktree isolation for pipeline commands (per-issue) (#999)
formalize shared state schema + Zod validation (#995)
Chat isolation foundation: worktree-gate + chat-registry integration (#996)
add xAI text-to-speech integration for on-demand text reading (#987)
interactive triage in Hermes task discovery (#984)
hybrid LLM+deterministic issue classifier (#982)
pipeline notifications to Telegram Pipeline topic (#981)
Telegram output quality for Maintenance (thread 33) (#980)
Telegram output quality for System Health (thread 29) (#978)
cron expansion — interactive commands, reaction chaining, and docs (#973)
route monitoring alerts and task discovery to forum health topic
pin opus versions and raise effort levels (#966)
add /jkz:new-chat command for parallel session isolation (#940)
surface retry count and per-call input tokens (#935)
Stage 0 confirmation UX — yes/no/adjust checkpoint after CAPM compute (reapply of #924) (#928)
Stage 0 confirmation UX — yes/no/adjust checkpoint after CAPM compute (#924)
run-research safer download-review-execute installer flow (#925)
add local sound notification on idle_prompt and permission_prompt (#919)
contradiction detection in knowledge_base before upsert (#916)
replace free-text deliverables with numbered menu in SCOPE phase (#913)
ask research type before topic in SCOPE phase (#910)
port extract-learnings skill from Claudest (#906)
agent calibration loop — /jkz:calibrate command (#904)
configurable penalty weight for false positives in feedback-loop (#903)
dynamic token budget for format-patterns based on remaining prompt space (#905)
port get-token-insights analytics skill from Claudest (#893)
YAML role metadata for validator agents (prerequisite for calibration) (#894)
port run-research skill from Claudest (yt-dlp inline) (#892)
port recall-conversations skill + memory-auditor / signal-discoverer agents (#861) (#896)
implement conversation transcript DB (claude-memory foundation) (#897)
research: derive CoE/WACC bottom-up as stage 0 (Analyst, not user input) (#882)
port improve-skill + create-agent from Claudest (#885)
port create-skill + repair-skill from Claudest (#883)
add response cache (TTL 1h) (#884)
invoke yahoo-finance/octagon/fred MCPs on-demand via Node proxy (#880) (#881)
Implement derivative research type in /jkz:research (#875)
add JKZ_CACHE_TTL_1H support for 1h prompt cache TTL (#856)
memory curation — 5-signal scorer and creation gate (#849) (#854)
chat-search tier-1 verify CBM index contents and freshness (#853)
Integrate codex-connector bot reviews into CR fix flow (#850)
CR: /chat context subcommand should surface amplio retrieval results (#844)
/chat subcommand UX — pin, response splitting, Markdown polish, full help (#843)
/chat session persistence — save/load/list/delete/export/search (#837)
semantic retrieval for /chat amplio context (#818) (#832)
add Polygon and QuantConnect MCP servers to research stack (#823)
/chat core engine — Haiku→Sonnet pipeline, write-through persistence, basic subcommands (#825)
test coverage for _ui/ server pipeline-reader and frontend hooks (#541) (#824)
CBM discovery + already-implemented check in /jkz:issue (#814)
doc-sync —apply mode — auto-update marked sections post-merge (#809)
DRACO query augmentation pipeline for /jkz:research (#573) (#806)
CBM auto-index staleness check — session start and pipeline commands (#807)
/jkz:ask + consultant role — direct model consultation (#805)
add agent YAML definitions and deploy script (#808)
hybrid search (vector + BM25/FTS5) for pattern retrieval (#799)
pause/resume CodeRabbit around Doctor commits (#802)
memory-curate.js — redundancy check + structure linter (#800)
Improve dependency diff accuracy: use npm audit fix —force —dry-run —json if available (#796)
add @cyclonedx/cyclonedx-npm as optionalDependency (#703) (#792)
/jkz:quality — expand to 13 categories + async/parallel execution + health report (#550) (#787)
outcome-based pattern scoring (#785)
supply-chain research type for /jkz:research (#782)
task-specific weighted rubrics for /jkz:research (#575) (#780)
ingest CR learnings into memory-store (#781)
Magic Docs — documentos auto-actualizados con markers (#772)
research pipeline — structured intake & output enhancements (#776)
mesh topology for jkz — read-only mesh + parallel gates + Byzantine voting (#641) (#765)
provider quota awareness — dynamic routing based on Codex quota state (#759)
attribution marker in assistant-posted GitHub comments (#728) (#755)
persist Doctor diagnostic reasoning between fix iterations via jkz:diagnosis-json signal (#752)
add —fix, —force, —preventive modes to deps-audit with MIGRATION_HINTS (#744)
smart error-type retry in wrappers (complement stall detection + rate-limit retry) (#738)
SCOPE enhancements — confidence scoring intake gate + source weights (A+B) (#743)
AUDIT/OUTPUT enhancements — contradictions + knowledge gaps (Features C+D) (#742)
Byzantine voting for Judge and Sentinel (2-of-3 majority) (#734)
Memory Dream — auto-consolidación de memoria con Haiku (#732)
add /jkz:ship ad-hoc QA + commit workflow (#735)
Governed Learning Layer Runtime Integration (#719)
verify pipeline PR merge state and issue closure (#725)
CR Pass Gate — formal QA gate for CodeRabbit findings (#717)
capture test gap findings and create jkz:ready issues post-merge (#723)
CR pipeline integration redesign — 3 checkpoints, thread resolution, quota control (#721)
Shadow Mode + Pattern Attribution (#666) (#711)
improve memory promotion criteria and integrate into /jkz:quit (#714) (#715)
generate implementation checklist on plan approval (#713)
Pattern Lifecycle Governance (#665) (#710)
security hardening — supply chain audit + compliance tiers (#700)
jkz:quit: check for unresolved CR findings on active pipeline PR before shutdown (#698)
make fallback effort and thinking configurable from .env (#696)
Merge + Post-Merge Outcome Tracking (#692)
Semantic git output compression in build-pr-diff.js (#691)
commands respect fallback_provider via resolve-provider-fallback.sh (#693)
MCP profiles — fix enableAllProjectMcpServers override + hardened toggle-mcp.js (#684)
Session Memory — resumen incremental de sesión con Haiku (#599) (#683)
dynamic compact instructions via InstructionsLoaded hook (#682)
staleness warnings for memory files (#674)
add AST-based large file context helpers (#661)
inject CR Plan into Auditor and Curator prompts (#654)
dynamic ground-truth injection for long-running agent loops (#653)
turn counter notification + lock purge (#655)

Fixes

pin compatible versions and add uv support to install script (#1088)
stdin prompt delivery and external model guards (#1082)
harden cleanup-merged with defense-in-depth safety guards (#1074)
STATE.json stale active_agent breaks Opus token tracking (#1076)
make CR pre-push CLI-primary and parse Type:potential_issue (#1078)
token_usage skips Opus subagents and review/QA wrappers (#1050) (#1058)
canonicalize hook server paths and warn on parallel-clone mismatch (#1059)
replace inline gh-token example with env-var pipe pattern (#1057)
hermes(docker-compose): remove unconditional docker.sock mount (#1053)
invoke context-safety hooks independent of cwd (#1037) (#1049)
prevent cross-host duplicate issues (#1046)
graceful exit when no git root in hook commands (#1033)
wire Telegram config + backfill script (#1029)
resolve entropy scan findings (issue #1021) (#1024)
remove npm cache from outcome-close workflow (#1016)
prevent CWD drift in all hook commands + node_modules symlink for worktrees (#1015)
detect all Blocked-by prerequisites + backfill jkz:blocked label (#1010)
route Telegram notifications to group topic (#1009)
enforce defensive patterns in infra scripts (#974)
headless pipeline invocation for Hermes container (#967)
extract verdict from response text in store-deliberation (#971)
update deprecated gemini-3-pro-preview in .env.defaults (#957)
use -p flag for non-interactive prompt delivery (#959)
replace deprecated gemini-3-pro-preview with gemini-3.1-pro-preview (#955)
CR follow-up on #941 — document —from flag and harden file copy (#946)
normalize terminal phases + coerce review_fix_count (#934)
recover verdicts missed by strict bold-only regex (#931)
refresh hook_executions on re-import (#930)
address post-port code quality issues in memory scripts (#899) (#926)
migrate polygon to mcp-massive v0.9.0 (rebrand) (#921)
quote descriptions containing unescaped colons (#918)
conform descriptions to Anthropic skill format (#917)
exclude worktree dirs from dry-check scan, replace em dashes (#872) (#911)
sync is_error on re-import for mid-flight tool calls (#909)
use .jkz-cbm-project-id anchor file for stable project name across mount paths (#890)
use portable path for .jkz permission entry (#889)
address out-of-scope findings from PR #843 (#851)
remove confirmation prompts from /jkz:quit
CI status check treats skipping/neutral as non-failure (#840)
chat-engine bugs and SLO string concat overflow (#836)
remove phantom MCP servers from RESEARCH_SERVERS and docs (#833)
inject PR body into review and QA agent prompts (#831)
resolve-cr-threads.sh —retry to handle REST-to-GraphQL propagation delay
warn when classify-issue confidence is not high (#821)
prevent double-count in cbm-index-check.js when timestamp is malformed (#811) (#820)
add projectId scoping to KNN-only pre-filter in queryPatterns (#813)
Telegram notification improvements — idle spam filter + enriched checkpoints + semantic routing (#812)
remove touch bash call from save.md to avoid permission prompt
add npm overrides to resolve transitive dep vulnerabilities (#798)
add npm overrides to resolve transitive dep vulnerabilities (#797)
smarter compliance floor, signal-based fallback, direct API call (#793)
register ignored attribution state on pipeline FAIL in feedback-loop.js (#794)
scope lockfile hash path by project in checkLockfileIntegrity (#791)
replace em-dashes (U+2014) with — in comments and strings (#789)
check circuit breaker after quota routing, not before (#775)
exclude worktrees dir from file collection (#788)
re-classify errors after context-overflow retry (#786)
detectTyposquatting returns skipped:true when assets file missing (#783)
wait for CodeRabbit to complete before fetching PR threads (#778)
early-abort on exit code 2 for non-retryable failures (#773)
create complexity labels before applying them to issues
write Judge checkpoint after wait in parallel mode (#768)
correct glossary findings from Codex audit (#760)
inject CURATOR_PLAN and AUDITOR_BLIND via printf after heredoc (#754)
execute qa.md steps 1-5.8 to include CR Pass Gate (#746)
improve classify-issue scope calculation and embeddings dimensions support (#737)
correct step-gate require call for cr_pass_gate
add || true to triage grep to prevent set -e failure on no match
fetch live PR state before /jkz:save to prevent stale context.md (#731)
sanitize —pr argument against path traversal in extract-test-gaps.js (#726)
support —delib-file flag in codex-wrapper and gemini-wrapper (#724)
resolve CR findings from PR #711 (#716)
enforce PR closing keyword verification in build and quick pipelines (#709)
verify OOS findings still exist before creating issue (#701)
emit project name at session start; remove velocity default cap (#694)
seed iteration 1 from existing CodeRabbit bot comments (#690)
add PR status check on session shutdown (#689)
resolve WSL/Windows path mismatch in test suites (#687)
portability and install bugs (#681)
portability and install bugs (#678)
fallback for import.meta.dirname undefined in esbuild bundles (#677) (#679)

Refactors

extract duplicated issue-type detection into shared skill (#1081)
address legitimate findings from entropy scan #1027 (1 stub + 2 test gaps) (#1047)
add tests for scripts/memory_lib (issue #953) (#1018)
fix entropy scan false positives (17 findings) (#1019)
split rate-limits.ts into focused modules (#1011) (#1017)
triage entropy scan false positives + add test coverage (#956) (#1000)
rename _ui/ to ui/ (#986)
replace em-dashes with — in source files (#923) (#929)
update Opus agent prompts for Claude Opus 4.7 behavior shifts (#922)
explicit layer ordering in prompt assembly (#908)
forward config excludePaths to dry-check collectFiles (#842)
uncertainty-based complexity classifier, unify trivial/quick/standard terminology (#839) (#855)
extract run_codex_exec helper in codex-wrapper.sh (#852)
separate Pipeline Judge (codex exec) from Codex PR Review (GitHub) (#845)
remove Phase 2 LLM call, return signals only (#829)
replace word tokenization with character trigrams in M1 classifier (#822)
fix entropy scan false positives and consolidate TEST_FILE constant (#804)
DRY timeline regex, flat-dir test detection, em-dash fixes (#774)

Performance

integrate context-mode MCP in research pipeline to reduce token accumulation (#741)
focused diff — cap total output, add prev-fix-file support (#733)
focused diff for Doctor on iteration 2+ (#647) (#688)
optimize model usage in CR triage and ambiguity gate (#675)

Other

bump actions/checkout from 4 to 6 (#1093)
bump actions/github-script from 7 to 9 (#1092)
bump softprops/action-gh-release from 2 to 3 (#1091)
bump actions/setup-node from 4 to 6 (#1090)
bump trufflesecurity/trufflehog (#1089)
disable cleanup-merged on session start (urgent, blocks #1055)
drop redundant top-level worktree-gate, switch in-session via EnterWorktree (#1052)
work-on-issue mandatory sequence rule (#1044)
cover research-output-gws.js and residual low-coverage files (#1004)
add component, page, and routing tests for ui/src (#952) (#997)
Add tests for get-token-insights/scripts (2090 LOC, from #936) (#998)
add tests for create-skill scripts + resolve TODOs (#950) (#988)
update classifier documentation for hybrid LLM+deterministic approach (#983)
add global positive-guidance rule (#975)
bump knip in the minor-and-patch group (#972)
align prose with ‘quick’ complexity tier nomenclature (#954)
branch coverage for monitoring-checks.js and research-capm-compute.js (#947)
cover feedback-loop and research-state CLI branches (#943)
cover pattern attributions and contradiction resolution (#941)
coverage for memory-curate branches and update-agent-calibration (#939)
cover session handlers, CLI, and main flow (#938)
coverage for deps-audit CLI branches and doc-sync scanner (#937)
bump the minor-and-patch group with 4 updates (#933)
bump typescript in /mcp in the minor-and-patch group (#932)
Revert “feat: Stage 0 confirmation UX — yes/no/adjust checkpoint after CAPM compute (#924)” (#927)
Adopt Opus 4.7 prompt patterns: severity, no-narration, default stance, Step 0/1 routing (#914)
document shallow-reasoning effort escalation for Opus 4.7 (#915)
clarify phase status and schema mapping in plan-ui-jkz (#912)
UI plans (jkz + ContentCurator) + python antipatterns reference (#902)
normalize .js/.cjs/.mjs to LF in .gitattributes (#878)
bump the minor-and-patch group with 2 updates (#871)
bump @types/node in /mcp in the minor-and-patch group (#870)
add derivative research type plan (#874)
clarify required vs optional research MCP servers (#835)
add /chat Telegram feature design plans (#815)
assert attribution state stays injected on PASS verdict (#795)
add retry_with_backoff tests covering early-abort on exit 2 (#779)
translate remaining Spanish strings to English (#771)
translate .claude/skills/ to English (zero regression) (#769)
translate global/commands/ to English (zero regression) (#767)
translate decisions, integrations, UI roadmap, and research templates to English (#766)
translate all project content to English (#764)
add glossary, pipeline-internals TOC, and ADR index in INDEX.md (#756)
navigation improvements (usage-guide, troubleshooting) (#753)
restructure agent-definitions.md with navigable headings (#751)
improve navigation and deep-dive internals (#747)
add enforcement note for FALSE_POSITIVE two-step workflow (#707)
allow Edit/Write on .claude/** and agents/** without permission prompt (#708)
daily artifact cleanup cron + gitignore test temp patterns (#705)
allow codebase-memory MCP tools and fix save.md write precondition
bump the minor-and-patch group in /mcp with 3 updates (#685)
bump the minor-and-patch group with 3 updates (#686)
Feat/notification and session duration (#676)
add honest escalation rule for Doctor iterations
add grounding skill for core opus agents (#672)
Add execution spine telemetry and handoff continuity flows (#668)
Feat/mcp profiles (#662)
bump dorny/paths-filter from 3 to 4 (#651)
bump actions/setup-node from 4 to 6 (#650)
bump actions/checkout from 4 to 6 (#649)
bump trufflesecurity/trufflehog (#648)

2026-03

Features

detect verdict_conflict in review and QA phases (#633)
add ReasoningBank RETRIEVE step (#636)
token consumption optimizations + CC-inspired improvements (#632)
away summary + agent progress (CC-inspired) (#622)
expand secret scanner with 5 new provider patterns (#591)
awaiting-input Telegram alerts + session duration tracking (#590)
saturation test for overly permissive rubrics (#589)
FTS5 trigram search in memory-store.js (#588)
hash-based cache invalidation for pattern summaries (#586)
auto-extract learnings on /jkz:quit via Haiku (#584)
dedup check in storePattern() via Jaccard similarity (#583)
skip-limit counter for auto-cleanup of stale patterns (#543) (#585)
integrate codebase-memory-mcp as structural knowledge graph (#582)
3-path context injection in format-patterns.js (#579)
rubric scoring with weighted criteria (DRACO formula) (#572)
auto-resolve CodeRabbit review threads after fixing (#571)
plan digest pre-compilation for downstream agents (#568)
enrich /jkz:issue with codebase context + architect evidence rule (#569)
file digest caching for agent context across iterations (#567)
reorder Builder prompt for server-side cache hits (#566)
deliberation caching by diff hash (#564)
CodeRabbit scan caching by commit SHA (#562)
pre-validated checks caching by diff hash (#561)
complexity classification fallback in pipeline entry points (#563)
add memory health warnings on session start (#554)
per-role provider fallback con cascada configurable (#540)
extend quality-scan with 5 new chore categories (#524) (#526)
subir cobertura de tests a 85% (statements, branches, lines) (#516)
mejorar informes Google Drive para research (Doc profesional + Sheets interactivas) (#514)
/jkz:commit — CR fix loop pre-push for ad-hoc commits (#501)
Visual Verification Pipeline — pass e2e screenshots to AI models (#499)
flaky test classification + formalized stop conditions in fix cycle (#498)
migrate research output to Google Workspace (gws CLI) (#500)
ontología de patterns + evidence tagging en research (#497)
phase timing + token tracking for dev and research pipelines (#490)
skill template based on TypeUI format (#495)
strengthen agent prompts — signal words, positional audit, model tendencies, MiniMax guidance (#488)
guard-destructive DB + Docker/K8s patterns (#421, #422) (#484)
Hardening hooks + CVE awareness + Smart Explore (#473)
align audit-permissions patterns with guard-destructive blocklist (#464)
Ollama lifecycle manager with reference counting (#460)
enhanced status line, context guardian, permission audit (#458)
memory curation skills + skill security auditor (#436)
Builder pre-push validators, repo structure map, entropy scan (#418) (#456)
two-phase degradation + role-aware strategies (#423)
headless AskUserQuestion via PreToolUse hook updatedInput (CC 2.1.85) (#413)
register jkz as Codex plugin for adversarial roles (Codex 0.117.0) (#411)
evaluate Codex sub-agents with path-based addressing for modular reviews (#412)
prioritize coderabbit:code-reviewer Agent in pipeline (#390)
Ambiguity Gate — proactive ambiguity detection at pipeline checkpoints (#386)
add /jkz:research command for financial research pipeline (#382)
issue complexity classifier with smart pipeline routing (#383)
research-output.js deliverable generator (#380)
templates + agent definitions para /jkz:research (#375)
add CodeRabbit local pre-push scan (step 4.5) (#376)
MCP finance servers + exceljs for /jkz:research (#374)
MiniMax M2.7 integration for adversarial roles (#362)
persistent store for researched ideas (#355)
apply Anthropic skill best practices (#354)
add context: fork to isolate heavy skills in subagents (#353)
add context: fork to isolate heavy skills in subagents (#350)
per-backend approver ID prompts in installer wizard (#338)
jkz update descarga desde GitHub + install limpio (#333) (#334)
clean install manifest + jkz update downloads from GitHub (#330)
harden jkz init/update (smoke test, atomic writes, tests, docs) (#325)
context-aware auto-save Stop hook (70%/85% thresholds) (#321)
fix pattern learning loop (#316)
improve save/load with rich context and per-chat snapshots (#317)
add project_id scoping to patterns table for per-project isolation (#318)
Haiku Transform Layer — plan-digest.js (gated, additive) (#300) (#313)
Spec Completeness Analysis — Architect Step 0 (32 dimensiones, adoption map, spec fidelity) (#305)
quality-scan —learn flag for pattern learning (#271) (#302)
optimización de costos — tier de modelos y fallback cross-provider (#292) (#299)
/jkz:quality — on-demand project quality scanner (#296)
bugs-scan.js — excluir test fixtures, allowPaths, async perf (#295)
integrar Context7 MCP en pipeline (pre-build, pre-fix, skill) (#290)
/jkz:bugs — Bug Finder en 3 niveles (Quick/Deep/Full) (#288)
Cross-Chat Awareness — chat registry + auto-context + /jkz:quit (#275) (#284)
Librarian Agent — knowledge broker centralizado (#276) (#285)
doc cache con fallback escalonado (stdin → cache → HTTP) (#282)
Session Survival — pm2 ecosystem config for persistent daemons (#278)
Deslop validator (ai-prose, diff-safe) (#283)
Project Scanner + Mapping (project-profile.json) (#281)
jkz update — sync frozen artifacts in plugin projects (#279)
AI traces validator + prompt injection defense + no-fabrication stability rules (#274)
robustizar skills con companion files y ${CLAUDE_SKILL_DIR} (#264)
auto-infer issue type from content in /jkz:issue and /jkz:start + fix qa.md detection (#265)
/jkz:e2e — agentic E2E UI testing (#257) (#263)
informed edges — evidence-based cross-role pattern affinity (#253) (#260)
skill discovery via /jkz:insights —skills (#255) (#261)
context inference — classify findings before asking in start/refine (#254) (#259)
hook server startup lock — atomic cross-platform duplicate prevention (#246)
adaptive KNN pre-filter with sqlite-vec for queryPatterns (#164) (#249)
LangFuse metrics UI for pipeline observability (#137) (#247)
declarative automations framework — final fixes (#142) (#250)
declarative automations framework — integrations (#148) (#242)
kanban board + goal tree — real-time pipeline view (#155) (#244)
cross-role pattern discovery via KIND_MAP (#163) (#241)
harden threat model against prompt-state poisoning (#240)
HTTP hook server for centralized observability (#152) (#215)
Beads design patterns — desire paths, section markers, agent mode (#169) (#238)
Experiment Journal + Post-Verdict Rollback (#168) (#236)
LangFuse UI integration — pipelines list, detail, metrics (#141) (#232)
add duration_ms to wrapper outputs for agent response time analytics (#216) (#235)
declarative automations — built-in actions (#147) (#231)
/jkz:resume — pipeline diagnosis and smart resumption (#223) (#233)
citation accuracy tracker en convergence-metrics (#167) (#230)
persistent threat model — cross-pipeline STRIDE accumulation (#151) (#218)
semantic routing para memory store (#162) (#210)
goal ancestry — GitHub sub-issues + pipeline state sync (#154) (#221)
verdict-blind prompting for validators (anti-groupthink) (#212)
trends and historical comparison for /jkz:insights (#157) (#220)
/jkz:issue — create issue from plan or text (#175) (#214)
event-driven CodeRabbit plan injection during planning phase (#209)
issue locking — prevent concurrent pipelines on same issue (#208)
echo behavior detector in convergence-metrics (#166) (#207)
error grouping integration in postmortem (#156) (#206)
analytics anti-bias improvements (#160, #161, #156, #165) (#205)
declarative automations framework core (#204)
JKZ_AUTO_APPROVE flag for pipeline checkpoint auto-approval (#202)
agreement rate monitoring (anti-groupthink) (#200)
auto-close parent issues when all children complete (#199)
inject fix_memory into Doctor prompt (#196)
Dependabot grouping, safety check, and CodeRabbit review (#189)
GitHub reaction-based approval polling for pipeline checkpoints (#184)
iterate CodeRabbit plan via Implement iterative refinement of CodeRabbit plans with feedback loop before Architect (#178)
LangFuse backend proxy with rate limiting (#145)
inject CR issue plans as Architect context (#139)
LangFuse observability tracing backend (#124)
stall detection, blocker eligibility, continuation turns, auto-cleanup (#122) (#123)
Mission Control sync adapter (#121)
web UI phase 2 — interactive chat via WebSocket + Agent SDK (#119)
web UI phase 1 — read-only session viewer (#115)
assemble-prompt centralization + heartbeat crash recovery (#107)
Kind.Agent classification, configurable retry, ultrathink, and CR fix loop (#105)
add /jkz:dev-self-review pre-PR consistency check (#106)
automate CHANGELOG and timeline generation (#104)
schema v4 — recency decay invariant + deliberation FTS (#81)
quality gates PR 1 — perf-quick + gitleaks + schedule integration (#87)
deduplicación cross-pipeline de findings (#85)
multi-stage clustering de findings para el Doctor (#84)
file locking completo en json-helper.js (#82)
add S1-S4 severity taxonomy to monitoring alerts (#83)
integrate web-quality-skills into jkz QA pipeline (#79)
CR reconciliation with OUT_OF_SCOPE + post-QA step (#72)
deep agents — 5 proposals for enhanced inter-agent communication (#71)
migrate shared contexts to native Claude Code skills (#70)
cross-phase feedback loop with root-cause analysis + smart maintenance (#69)
cross-phase feedback loop with structured root-cause analysis (#66)
step-gate enforcement for pipeline step sequencing (#68)
add branch discipline rule for ad-hoc work (#63)
pipeline improvements from codepen-extractor retrospective (#62)
context management + pattern learning loop (#59)
issue type-aware pipeline (bug/refactor/feature) (#57)
CLI version freshness and /jkz:health command (#58)
remote installer — curl | bash zero-clone setup
refactor jkz CLI — install (global) + init (per-project)

Fixes

bash 3.2 compat + expand eslint/prettier config detection (#634)
reset degraded_response and stale_deliberation flags on clean runs (#639)
use —git-common-dir for shared state dir across worktrees (#640)
debounce context-monitor hooks + /jkz:save hint in CRITICAL (#635)
add semantic prompt injection patterns to guard-injection (#627) (#631)
deduplicate entropy-scan findings (#578) (#593)
resolve 6 pre-existing test failures (#592)
reduce entropy scan false positives (#523) (#576)
resolve OUT_OF_SCOPE findings from PR #564 (#565)
CRLF scanner only reports shell scripts (291 warnings eliminated)
pass allowPaths to TODO scanner, eliminating 36 false positives
include .md files in config orphan scan
harden entropy-scan dedup to prevent issue spam
gws-client Windows compatibility and real API commands (#503)
entropy-scan dedup — worktree-safe repo name + local state file (#483)
codex plugin setup script Windows compatibility (#415)
migrate auth from localStorage to httpOnly cookies (closes #401) (#409)
batch CR findings (5 issues) (#404)
use find -L to follow symlinks in smoke test (#403)
allow hyphens in role names and update plugin/hook paths (#402)
migrate from pandoc to native Node.js libs (docx, pptxgenjs, pdfmake) (#395)
portabilize MCP finance paths and document FRED_API_KEY (closes #391) (#392)
resolve npm audit vulnerabilities (picomatch, smol-toml, yaml) (#389)
handle array data, dynamic sheets, auto-resolve reference docs (#388)
missing slash in jkz.cmd path conversion
use git clone instead of tarball for persistent updates
register research agents in resolve-wrapper + clarify output sections (#379)
adapt parseKnipOutput for knip v6 JSON format (#369)
bump flatted 3.4.2 (GHSA-rf6f-7fwh-wjgh) (#365)
ensure pipeline closes issue and cleans agent labels on completion (#357)
resolve 24 pre-existing test failures (hook-server, metrics-db, jkz-update) (#356)
allowlist .claude/context.md in guard-worktree
pass userId to dispatchCommand (was referencing undefined msg) (#341)
plan checkpoint displays plan content before asking for approval (#336)
resume delegates to /jkz:pipeline —resume instead of standalone commands (#335)
replace inline approval polling with dedicated script + sentinel recovery (#323)
changelog-review & insights — CodeRabbit detection, Haiku spawn, VSCode tab prevention (#315)
align spec fidelity density gate across Architect, Auditor, and Curator
parse Codex web changelog (developers.openai.com HTML format) (#308)
auto-unblock issues on merge via GitHub Action (#307)
repair 2 broken test files (#303)
dependency-reconcile.js — auto-unblock legacy blocked issues (#293)
dependency-reconcile.js — auto-unblock legacy blocked issues (#294)
use GitHub auto-close keywords (Closes/Fixes) instead of Implements in PR body
sweep orphaned worktree directories on disk during session-start cleanup
parent issues no se desbloquean cuando sub-issues se completan (#248)
auto-register body dependencies before check-deps (#239) (#243)
parse new Gemini CLI JSONL format (#222) (#237)
auto-close issues referenced in merged PR commit message (#217)
github-review.js does not support —stdin-file argument (#201)
add retry logic to validate-phase.sh for GitHub API eventual consistency (#203)
/jkz:quick no reconcilia comentarios inline de Judge/CodeRabbit (#198)
remove invalid &; syntax from pipeline-watcher launch (#171) (#195)
isolate codex-wrapper test state to prevent flaky failures (#191) (#194)
detect dead background tasks via heartbeat + PID tracking (#185)
auto-cleanup merged worktrees at session start (#188)
defensive cwd for gh commands in plugin mode (#181)
resolve-wrapper CWD inheritance + migrate mktemp to state/tmp (#182)
harden callHaiku() env stripping and stdin handling (#176)
resolve /dev/stdin ENOENT on Windows and Codex thumbs-up false FAIL
resolve GHSA-46wh-pxpv-q5gq + add Dependabot (#128)
close web UI phase 0 test and auth gaps (#118)
update minimatch to patch high-severity ReDoS CVEs
update hono and @hono/node-server to patch high-severity CVEs
switch gitleaks dir → detect —no-git (fixes exit 126) (#103)
implement finding lifecycle resolve on PASS via persisted hashes (#102)
correct secrets-scan output parsing and sed redaction false positive (#100)
prevent TOCTOU race in stale lock removal (#99)
async withFileLock and generic drive letter in jkz.cmd (#98)
add -v flag and allowlist SHA256 checksums (#96)
scope c8 to project files with —include flag (#95)
renumber schema migration to v5 to prevent collision with feat/memory-enhancements (#93)
FINDING_HISTORY siempre vacío — bash pipe subshell en fix.md (#92)
hoist require() ordering and fix multiline PR comment parsing
add explicit AskUserQuestion options to prevent merge-implying labels
installer robustness — auto-repair, version checks, pipefail safety
installer bugs — wizard options, banner z, junction, gemini model
PowerShell/CMD support for jkz CLI on Windows

Refactors

add path-scoped frontmatter to .claude/rules/ (#625) (#630)
fix ai-prose, ai-traces, and dry-check findings (#578) (#594)
separate plan-digest output paths by mode (compile vs summarize) (#581)
deduplicate shared constants (BOT_LOGIN, PRESCAN_MARKER, BLOCKED_BY_RE) (#507) (#517)
remove legacy research output stack (#502) (#509)
reduce entropy scan false positives (#489) (#505)
replace em dashes and AI prose patterns (#481) (#485)
template-based generation for research deliverables (DOCX/PPTX/PDF) (#463)
slim CLAUDE.md + rules (~50% token reduction) (#462)
block access to credential paths (#420) (#459)
adopt gh 2.88.x flags (—exclude, —duplicate-of, —blame) (#346) (#364)
document Codex 0.116 view_image capability in adversarial agents (#347) (#363)
CC hooks StopFailure + CRLF eval (#342) (#360)
document Gemini 0.34.0 flag decisions (#345) (#359)
plugin mode CLAUDE_PLUGIN_DATA + PLUGIN_SEED_DIR múltiple (CC 2.1.78-79) (#358)
add CC 2.1.78 agent frontmatter fields (effort, maxTurns, disallowedTools) (#351)
resolve 57 knip findings (stale cleanup) (#331)
estandarizar skills — naming, frontmatter y progressive disclosure (#258) (#266)
reorganize docs, skills, and add phase-log tracking (#88)
remove redundant issue approval in /jkz:start

Performance

pass compact-plan JSON to Curator instead of full verbatim plan (#637)
reduce context overhead — autocompact 70%, slim CLAUDE.md, compact instructions (#521)
Context7 plan-specific queries + cache docs for fix iter 2+ (#520)
reduce Claude quota usage across pipeline (#519)
pipeline cost optimizations — caching and model routing (#515)
add context:fork to qa-security-scan, memory-review, skill-security-audit skills (#487)
reduce agent effort levels for pipeline redundancy (#482)
Optimize querySimilarDeliberations with KNN pre-filter (#320)
GPT-5.4 integration + model-aware thresholds for 1M context (#117)
optimize GitHub Actions to reduce ~85% monthly minute usage (#116)
background post-processing in wrappers (#114)
robust feedback loop with fuzzy matching (#111)
add cross-phase PR diff and metadata cache (#113)
externalize raw output from deliberation JSON (#112)
gate compress/summarize by prompt byte size (#110)
dedupe pattern injection — remove manual patterns superseded by assemble-prompt middleware (#109)

Other

test+fix: global/scripts coverage and ai-prose self-exclusion (#578) (#597)
reframe README — multi-model rationale, command graph, mermaid diagrams (#580)
update README with new roles, commands, and features (#577)
subir cobertura de 81.65% a 90.13% (#547)
remove unused adm-zip dependency and fix em dash in test (#522)
update GWS integration docs post-PR #514 (#518)
bump typescript from 5.9.3 to 6.0.2 in /mcp (#513)
bump knip in the minor-and-patch group (#512)
bump @modelcontextprotocol/sdk (#511)
add git stash clear, reflog expire, filter-branch patterns (#443)
investigation report for Gemini 0.35.x AgentLoopContext (#408) (#410)
add research pipeline, MCP servers, complexity classifier, fix evidence-hierarchy path (#393)
bump knip from 5.86.0 to 6.0.1 (#367)
bump the minor-and-patch group with 2 updates (#366)
document Gemini Tracker spike + @copilot enrichment evaluation (#348) (#361)
configure prettier, eslint, and quality npm scripts (#340)
quality audit + coverage improvements for 10+ files (#328) (#332)
update documentation accuracy (model versions, metrics, skill names, commands) (#327)
add prettier, eslint, knip + fix quality-scan Windows compat (#310)
fix flaky tests — doc-cache Jest→node:test, orchestrate lock isolation, codex-wrapper EBUSY retry (#306)
bump the minor-and-patch group in /mcp with 2 updates (#298)
bump better-sqlite3 in the minor-and-patch group (#297)
deslop — replace em dashes (U+2014) across 216 files (#301)
standardize issue dependency format and auto-labeling rule (#280)
Chore/claude code 2.1.75 settings (#252)
adopt Claude Code 2.1.75 settings (#251)
add subprocess coverage for cr-plan-iterate main() loop (#190)
bump @types/node from 20.19.33 to 25.4.0 in /mcp (#193)
bump trufflesecurity/trufflehog (#192)
bump actions/upload-artifact from 4 to 7 (#133)
bump actions/github-script from 7 to 8 (#132)
bump actions/checkout from 4 to 6 (#131)
bump @modelcontextprotocol/sdk in /mcp (#136)
bump c8 from 10.1.3 to 11.0.0 (#130)
bump actions/setup-node from 4 to 6 (#129)
add implicit approval coverage for normalize-judge-verdict
LangFuse docs, UI roadmap, and installer labels (#126)
pause web UI, rename to _ui/ (#120)
run tests only on push to main, not on PRs
remove legacy skill dirs and apply minor doc fixes
Merge branch ‘main’ of https://github.com/j0KZ/jkz_Multi-Agent_System
Merge branch ‘main’ of https://github.com/j0KZ/jkz_Multi-Agent_System
Merge branch ‘main’ of https://github.com/j0KZ/jkz_Multi-Agent_System
Merge branch ‘main’ of https://github.com/j0KZ/jkz_Multi-Agent_System
Merge branch ‘main’ of https://github.com/j0KZ/jkz_Multi-Agent_System
Merge branch ‘main’ of https://github.com/j0KZ/jkz_Multi-Agent_System
Merge branch ‘main’ of https://github.com/j0KZ/jkz_Multi-Agent_System
Merge branch ‘main’ of https://github.com/j0KZ/jkz_Multi-Agent_System
increase test coverage to 90%+ (#80)
Merge branch ‘main’ of https://github.com/j0KZ/jkz_Multi-Agent_System
Merge branch ‘main’ of https://github.com/j0KZ/jkz_Multi-Agent_System
Chore/test coverage 86 (#74)
Merge branch ‘main’ of https://github.com/j0KZ/jkz_Multi-Agent_System
Merge branch ‘main’ of https://github.com/j0KZ/jkz_Multi-Agent_System
Merge branch ‘main’ of https://github.com/j0KZ/jkz_Multi-Agent_System
Feat/start refinement dry (#64)
add detailed phase explanations and Orchestrator role to README
update README with new pipeline steps, commands, and infrastructure
Feat/context management pattern learning (#60)
update ASCII banner to bold blocks for better readability
update local Bash permission patterns for installer commands
add API backend mention and updated CLI commands to README
add npm link global CLI installation instructions
update all documentation to reflect PRs #52-#55

2026-02

Features

hybrid Judge, per-role model, ADR/STRIDE/SLOs/deps-audit (#55)
ADR extraction, STRIDE threat modeling, pipeline SLOs, postmortem, deps audit (#54)
security hardening, statusline, SQLite metrics, MCP sessions (#52)
pipeline export, shared retry, doctor auto-fix, deep health checks
add stub validator, premortem risks, deslop and perf-audit commands
Docker CLI isolation — all CLIs native, no API routing (#49)
add jkz CLI entrypoint with Windows support
plugin architecture with dual-root path resolution (#43)
expand MCP server with new tools and resources
add hook-runner.js with JSON config and update all hooks
centralized env loading and MCP launcher
add toggle-plugins CLI with profiles and plugin manifest
changelog insights with Haiku analysis + CodeRabbit tracking
reject-with-feedback flow in Telegram bot + add lens to Gemini routing
add semantic memory system (embeddings + pattern scoring + fix-memory)
add cross-role pattern accumulation and structured fix context
add Chair Test rule to Architect prompt
add CodeRabbit Issue Planner configuration
add Phase 3 integration tests and telegram-bot refactor
add PR auto-review workflow, switch to OAuth token, remove invalid —max-turns flag
implement GSD-inspired improvements across pipeline
add unit tests for core components (Phases 1-2)
add calibration, scope clarification, and programmatic validators
add monitoring loop, task discovery, and prompt rewriting
add CI feedback workflow and escalation chains
normalize codex review output for pipeline compatibility
add GC for closed issues in STATE.json
add cortex briefing, file ingestion, and autonomy levels (SpaceBot round 2)
add natural language intent detection via Claude Haiku CLI
add PostToolUse hooks, pinned comments, stream-json, MCP resources, Codex memory, merge-gate check, Projects sync
add Check Runs API, minimize outdated comments, agent frontmatter
cost tracking, memory digest, watcher circuit breaker
quick wins ronda 2 - Notification/PreCompact hooks, language, spinnerVerbs, rules/
add message coalescing for rapid free-text input
Doctor fix-signal auto-governance of loops (#36) (#40)
pre-allocate comment slots for parallel QA/Sentinel execution (#39)
multi-project Docker setup + pipeline housekeeping
add review checkpoint (checkpoint 1 of 2 in pipeline loop)
add QA human checkpoint as 3rd pipeline gate
add pipeline watcher, harden Judge/Inspector calibration
add merge gate health checks to health-check.sh
add init-merge-gate.sh for cross-project merge protection
add auto-revert for unauthorized PR merges
add server-side merge gate — human-only PR approval
lock on create, unlock on clean
add _parse_worktrees() helper for —porcelain -z output
add —check-updates flag for CLI version freshness checks
add /jkz:refine command for pre-plan issue refinement
integrate CodeRabbit as enrichment step in review phase (Step 2.5)
per-agent token tracking with summary table in notifications
per-role temperature config + benchmark script
Telegram bot daemon for pipeline control from phone
stop reaction + pre-seeded checkpoint emojis
automatic model fallback for all LLM wrappers
security hardening S7 — Gemini web search off, read-only mode, optional sandbox
add unit tests for json-helper.js + wire into CI monitor
security hardening S6 — disable web search + sandbox Judge (codex review)
output quality Phase 3 + security hardening S1-S5
token efficiency Phase 1+2 — system/user split, verdict signals, diff filtering, shared signals DRY
structured Discord/Slack notifications — TL;DR + link
collapsible PR comments, checkpoint notifications, orchestrator TL;DR
resilient polling with sentinel files for session-reset recovery
Discord threads + sentinel state tracking for agent runs
activate unused Claude Code features (hooks, effort, CoT, TodoWrite)
add jkz:regression label to labels.yml and install.sh
add regression issue template for manual reports
add continuous monitoring workflow and script
add Discord reaction polling to E2E test
add Discord bot-only mode (no webhook required)
add bidirectional feedback, file locking, bash 3.2 compat, and E2E test
add universal API wrapper for OpenAI-compatible backends
add cross-issue dependency tracking with cycle detection (roadmap item 6)
add issue decomposition — complexity check in plan phase (roadmap item 4)
add Slack/Discord notifications, autonomous pipeline, and R2 audit fixes
add deliberation analytics with phase/command schema enrichment (roadmap tier 2)
add fetch-md.js helper for web content pre-cleaning
add agent issue proposals and post-merge workflow (roadmap tier 1)
add CLI version reporting to health-check.sh
add update-readme-status script (closes #1)
add global commands/scripts mirror and install script
add checkpoint system, guard hooks, phase validation, and reference docs
formalize orchestrator role with agent definition and state machine
rewrite agent personalities with distinct voices and identities

Fixes

use $@ for forward-compatible arg passing and error on unknown commands
replace MIT license with proprietary notice (#46)
address CodeRabbit review findings (#44)
address CodeRabbit review findings
strip all CLAUDE* env vars for nested CLI invocations
use spawnSync with temp file and unset CLAUDECODE for nested sessions
burst protection and kill switch to stop notification spam
timestamp parsing, recency filter, and auto-update CLIs
rate-limit Telegram notifications and fix emoji encoding
disable real Telegram notifications in orchestrate tests (#41)
add /usr/sbin to launchd PATH for sysctl dependency
address QA findings (Lens + Sentinel pass)
address code audit findings (Codex pass)
5 critical bugs and quick wins
finish Lens rename in test script and docs
add statuses:write permission to merge gate workflows
warn on stderr when git worktree list fails in _parse_worktrees
clarify main-tree serialization in Rule 6
make git worktree lock best-effort in create_worktree()
scope Rule 6 to per-issue with worktree isolation
improve bot free-text response + document launchd TCC issue (#22)
worktree cleanup detects squash merges + MCP server tests
resolve 7 findings from first E2E pipeline run (#18)
cross-platform compat in health-check.sh
resolve 7 integration findings from first E2E pipeline run
add LICENSE commit to CHANGELOG (Judge review P2)
project audit — macOS compat, security hardening, docs sync, setup automation
pass GH_TOKEN to health-check.sh and surface issue creation errors
add private-key / PRIVATE_KEY redaction patterns (QA iter 3, Doctor final)
add Basic auth pattern and use .* for multi-word value redaction (sentinel qa3)
add x-api-key redaction patterns and cover api-key hyphen form (QA iter 2)
extend sed redaction to cover key=value and key: value forms (sentinel qa2)
isolate GH_TOKEN, sanitize issue output, harden eval (QA iteration 1)
replace eval with bash -c and sanitize issue output (sentinel qa)
add safe defaults for GitHub env vars (set -u compat in local runs)
bash 3.x compatibility for health-check.sh and api-wrapper.sh
apply audit findings — worktree exactness, QA worktree mode, plan contract
filter worktree cleanup by branch name, not path
resolve pipeline friction from first E2E run
escape JSON values and fix pipefail+head SIGPIPE in health-check
add explicit prompt templates to plan.md and clear stale README status
harden worktree guard with protected-directory denylist and require Doctor comment in fix validation
harden guard hooks, phase validation, and arithmetic bugs after 3-pass Codex audit
add Node.js launcher (run.js) to fix broken bash builtins in sandbox
replace jq with Node.js json-helper, add —prompt flag, fix sandbox CLI checks
add bash prefix to all script invocations
improve install.sh chmod messaging for Windows
harden gemini-wrapper.sh for Windows compatibility
harden codex-wrapper.sh + add Judge codex review path
reimplement MCP health_report in native Node.js

Refactors

rename QA agent to Lens
rewrite list and cleanup-merged with callback parser

Other

Feature/security statusline metrics sessions (#53)
Merge branch ‘main’ of https://github.com/j0KZ/jkz_Multi-Agent_System
Feature/docker cli isolation (#51)
Feature/notification redesign (#48)
Merge branch ‘main’ of https://github.com/j0KZ/jkz_Multi-Agent_System
Feat/plugin installer (#47)
Merge branch ‘main’ of https://github.com/j0KZ/jkz_Multi-Agent_System
comprehensive documentation overhaul (#45)
Merge branch ‘main’ of https://github.com/j0KZ/jkz_Multi-Agent_System
update local permissions and tool settings
improve wrappers, add utilities, env integration
improve pipeline commands and skill definitions
add .worktrees/ to .gitignore
round 4 coverage to 93.25% with 490 total tests
Fix/test notify spam (#42)
round 3 coverage to 81% with 54 new tests (416 total)
add 6 test files for round 2 coverage (hooks, utils, analytics, checkpoint)
add 9 test files (49 tests) for validators, hooks, and utilities + c8 coverage
add Claude DevTools installation guide (Windows, macOS, Docker)
add v0.1.4 changelog entry for ronda 2 features
mark Windows as tested in notifications platform table
Revert “feat: Doctor fix-signal auto-governance of loops (#36) (#40)”
Revert “feat: pre-allocate comment slots for parallel QA/Sentinel execution (#39)”
Revert “Fuzzy repair en parsing de verdict JSON (#38)”
Fuzzy repair en parsing de verdict JSON (#38)
Revert “Add plan-diff monitoring to Build→Review handoff (#34)”
Add plan-diff monitoring to Build→Review handoff (#34)
Reinforce anti over-eagerness instructions in Builder/Doctor prompts (#33)
document merge gate across CLAUDE.md, setup-guide, and SELF-REVIEW
undo PRs #30 and #32 — merged without QA phase
Git worktree upgrades — Parallel Builders: state migration + guard hooks (#31)
Add plan-diff monitoring to Build→Review handoff (#32)
Reinforce anti over-eagerness instructions in Builder/Doctor prompts (#30)
Merge pull request #29 from j0KZ/jkz/issue-22
Merge pull request #28 from j0KZ/jkz/issue-24
update header with cleanup-merged, lock, and git version note
add worktrees comparison section to usage guide
add .claude/worktrees/ to .gitignore
add usage guide reference to README
add usage guide with workflows, commands, and tips
sync documentation with Feb 22 features (CHANGELOG 0.1.3, CodeRabbit, Telegram, token tracking)
Revert “feat: per-role temperature config + benchmark script”
Merge pull request #19 from j0KZ/jkz/issue-18
add CHANGELOG v0.1.1 — model fallback, E2E fixes, cleanup
Topic 10 nice-to-have — model upgrade, cleanup, test docs
Merge pull request #17 from j0KZ/jkz/issue-16
add OrbStack as Docker alternative, troubleshooting for macOS osascript bug
add CHANGELOG.md for v0.1.0 (72 commits, Feb 17-21)
add S6 security hardening to README table and setup guide
sync documentation to latest commits (Feb 19-20)
Merge pull request #13 from j0KZ/jkz/issue-12
allow notify.sh bash invocations in local settings
mark roadmap item 5 (Continuous Monitoring) as COMPLETED
add state/monitor-report.json to .gitignore
update Gemini model to 3.1-pro-preview and read models from .env
update notifications and roadmap for Discord bot-only mode
update CLAUDE.md for universal API wrapper architecture
add npm and install permissions to settings
update README and roadmap for deliberation analytics (roadmap items 1-3 complete)
apply 14 iterate audit resolutions to roadmap
add setup guide and auto-improvement roadmap
Merge branch ‘main’ of https://github.com/j0KZ/jkz_Multi-Agent_System
Merge pull request #3 from j0KZ/jkz/issue-2
add MCP server documentation, troubleshooting guide, and update README structure
sync README structure diagram and checklist with actual project state
add pipeline test results (codepen_extractor)
update README with bash prefix + mark e2e test complete
update CLAUDE.md with bash prefix + Judge quota note
Add LICENSE, fix broken README links, add repo metadata
Initial commit: jkz multi-agent development system