Skills

A skill is a SKILL.md file under .claude/skills/. Each one is a focused capability the system can load on demand: some are injected automatically into agent prompts, others are invoked by a slash command, and a handful trigger from natural-language phrases.

This page catalogs all 39 skills. Two columns matter for every entry:

Purpose — what the skill does, in one line.
Trigger — how it activates. For internal skills this is where it is injected; for invocable skills it is the phrase or command that loads it.

Internal vs. invocable. Framework and most QA/model skills are internal: the pipeline injects them automatically and you never call them by name. Skills marked (user-invocable) can be triggered directly by a phrase or a /jkz:* command.

Framework skills

Internal skills that define the system’s base behavior. Injected automatically into agent prompts — never invoked directly.

Skill	Purpose	Injected in
`framework-stability-rules`	Stability rules: no repeating rejected approaches, evidence required, explicit deltas per iteration, prompt-injection defense, no fabrication.	All agent prompts
`framework-signal-format`	Shared signal formats (Context Protocol, Issue Proposal, Iteration Signal / `verdict-json`) for inter-agent communication.	All agent prompts
`framework-evidence-hierarchy`	Evidence hierarchy for evaluating arguments: Execution > File Citations > Reasoning.	Adversarial roles (Auditor, Judge, Sentinel)
`framework-agent-grounding`	Grounding and failure-discipline: no invented references, trimmed process narration, fail-fast reporting.	Architect, Builder, Doctor
`framework-coderabbit-context`	CodeRabbit integration context for enrichment review (non-gating pre-scan).	Judge, Inspector (review phase)
`framework-issue-type-detection`	Issue type classification (bug / feature / refactor / chore) by keyword matching, with an `AskUserQuestion` fallback.	`/jkz:issue`, `/jkz:start`
`plan-stride-analysis`	STRIDE threat-modeling template (Spoofing, Tampering, Repudiation, Information Disclosure, DoS, Elevation of Privilege).	Architect, Sentinel — conditional on security labels or ≥2 security keywords

QA skills

Quality and security checklists, injected during the QA phase based on the PR’s content (frontend, backend, compliance tier).

Skill	Purpose	Trigger
`qa-web-a11y`	WCAG 2.1 accessibility checklist (POUR, contrast, keyboard, ARIA, screen readers).	Injected: Lens (QA, web-renderable diff)
`qa-web-performance`	Core Web Vitals and performance budgets (JS/CSS/total size).	Injected: Lens (QA, web diff)
`qa-web-best-practices`	HTTP security headers, CSP, input sanitization, secure cookies.	Injected: Sentinel (QA, web diff)
`qa-compliance-basic`	Tier 1 compliance: dependency pinning, committed lockfile, license, basic security.	Injected: Lens / Sentinel (compliance tier ≥ 1)
`qa-compliance-regulated`	Tier 2 compliance: audit trail, PII handling, data residency, SOC2 / HIPAA controls, SBOM.	Injected: Lens / Sentinel (compliance tier ≥ 2)
`qa-security-scan`	Red/Blue team security assessment in a single session.	Injected: Sentinel (QA); also on-demand pre-deploy
`qa-web-quality-audit`	Full Lighthouse web quality audit (Performance, Accessibility, SEO, Best Practices).	(user-invocable) “audit my site”, “web quality”, “check Lighthouse”

Model router skills

Skills that route a prompt to a specific model or model class.

Skill	Purpose	Trigger
`model-codex-ask`	Route a prompt to the configured adversarial backend for adversarial analysis.	”ask codex”, “codex review”, “adversarial check”, “audit with codex”
`model-gemini-ask`	Route a prompt to Gemini for validation analysis.	”ask gemini”, “validate with gemini”, “gemini check”
`model-debate-plan`	Structured adversarial debate between 2–3 models, each defending a position.	Used internally by `/jkz:debate`
`model-round-robin-ask`	Serial multi-model consultation where each model sees prior responses.	Used internally for the Architect → Auditor → Curator flow
`model-context7-lookup`	Look up external library documentation via Context7 MCP.	(user-invocable) library / framework / SDK API questions, version migration, CLI usage

Utility skills

Everything else: authoring, memory, docs, research, and the idea vault. Most of these are user-invocable.

Skill & agent authoring

Skill	Purpose	Trigger
`create-skill`	Author a new skill or slash command.	”create a skill”, “make a command”
`create-agent`	Author a new subagent (system prompt + frontmatter).	“create an agent”, “build a subagent”, “design an autonomous agent”
`improve-skill`	Add features or improve an existing skill’s effectiveness.	”improve a skill”, “make this skill better”, “what’s missing from this skill”
`repair-skill`	Diagnose and fix structural issues in a skill.	”fix my skill”, “audit this skill”
`skill-security-audit`	Scan skill files for security patterns (code/prompt injection, data exfil, jkz violations, secrets).	“audit skill”, “skill security”, “scan skills” (`/jkz:skill-audit`)
`dev-self-review`	Pre-PR self-review for jkz integration consistency.	”self review”, “check my changes”, “review before PR”

Memory

Skill	Purpose	Trigger
`memory-review`	Review memory files for staleness, overlaps, and promotion candidates.	”review memory”, “stale memories”, “memory cleanup”
`memory-promote`	Score a memory file for promotion to a rule or CLAUDE.md section.	”promote memory”, “memory to rule”, “should this be a rule”
`memory-status`	Memory health dashboard (status, counts by type, recent changes).	“memory status”, “memory health”, “memory dashboard”
`extract-learnings`	Persist learnings to memory or maintain existing memories.	”extract learnings”, “remember this pattern”, “consolidate memories”, “dream”

Docs & content

Skill	Purpose	Trigger
`make-changelog`	Generate or update `CHANGELOG.md` (Keep a Changelog 1.1.0) via parallel Haiku subagents.	”update the changelog”, “regenerate changelog”
`update-readme`	Propose `README.md` updates via parallel Explore subagents, behind a human approval gate.	”update the README”, “refresh README.md”
`update-claudemd`	Refresh a stale CLAUDE.md and reconcile diverged `rules/` files.	CLAUDE.md grown stale, `rules/` exceed ~300 lines, content needs promotion/demotion
`wiki-generator`	Generate or refresh the public jkz wiki at jkz-docs from the private repo.	”wiki-generator”, “regenerate wiki”, “publish wiki” (internal pipeline)
`convert-to-markdown`	Convert a webpage URL to clean markdown (EzyCopy CLI).	“convert this URL to markdown”, “scrape page as markdown”, “ezycopy”
`repo-to-markdown`	Convert a GitHub repo or local directory to a single structured markdown doc (repo2md).	“read this repo”, “convert repo to markdown”, “repo2md”, a raw GitHub URL

Research & recall

Skill	Purpose	Trigger
`run-research`	Research a topic or surface what’s trending and being discussed recently.	”research a topic”, “last30”, “what’s happening with X”, “latest on X”
`recall-conversations`	Recall, search, or continue past conversations.	”what did we discuss”, “continue where we left off”, “remember when”
`get-token-insights`	Analyze Claude token usage, cache hit rates, and cost optimization.	token usage, spend, cache hit rate, or cost questions

Idea vault

Skill	Purpose	Trigger
`vault`	Manage the persistent idea store: save, list, search, and mark ideas done.	”save to vault”, “what’s in the vault”, “ideas backlog”, “save this for later” (`/jkz:vault`)

References

CLI / commands — the /jkz:* slash commands that drive these skills.
Architecture — how skills fit into the three-phase pipeline.