skill-publish
Publish a skill after validation and approvals; updates the catalog and logs the publish event.
- License 许可
- AGPL-3.0
- 最近更新
- 2026-08-15
- Stars
- ★ 30
- Forks
- 7
01 前言 Front Matter
- name
- skill-publish
- description
- Publish a skill after validation and approvals; updates the catalog and logs the publish event.
- allowed-tools
- Read, Grep, Glob, ApplyPatch, Write
- argument-hint
- [skill-name]
- disable-model-invocation
- true
02 资源文件 1
- 参考文档manifest.yaml594 B
03 SKILL.md 正文
Skill Publish — Controlled Release
Purpose
Publish a validated skill with proper approvals, then update catalog metadata and log the publish event.
Inputs
-
Skill name
-
Evidence of test gates
-
Explicit approver (required for high blast radius)
-
Template paths are resolved from
${AGENTS_SKILLS_ROOT}/skill-publish/.
Steps
- Run
/skill-validate <skill-name>and confirm PASS. - Enforce approval gate
- If
blast_radius: high, require explicit approval from an authorized maintainer. If missing, STOP.
- If
- Update catalog entry
- Increment version if needed
- Update
updateddate
- Write publish log to
.jarvis/context/skills/publish-log.mdwith:- Skill name, version, owner, blast radius
- Approver (if required)
- Date and summary
- Return confirmation
Decision Trace Protocol
This skill participates in the skill evolution system by capturing decision traces at gate resolutions and consulting accumulated feedback.
Trace Consultation (short loop)
Before executing any step with a quality or human gate, query accumulated decision traces:
python3 "${AGENTS_SKILLS_ROOT}/_shared/trace_query.py" recent \
--skill "skill-publish" --gate "<gate_name>" --limit 5 --min-loops 1
If patterns or recent feedback exist, incorporate them as additional constraints. Do not cite traces to the user unless asked.
Trace Capture (after gate resolution)
After every gate resolves, capture a decision trace:
python3 "${AGENTS_SKILLS_ROOT}/_shared/trace_capture.py" capture \
--skill "skill-publish" \
--gate "<gate_name>" --gate-type "<human|quality>" \
--outcome "<approved|rejected|passed|failed>" \
--refinement-loops <N> \
[--feedback "<user's feedback text>"] \
[--category <CATEGORY>]
Post-Run Retrospective
After completion, ask: "Any feedback on this skill-publish run? (skip to finish)" If provided, capture via trace_capture.py with gate "run_retrospective" and gate-type "retrospective".
Output
- Confirmation message
- Link to catalog entry + publish log
04 同仓库更多 Skills
wiki-research
Autonomous research orchestrator for Jarvis Wiki domains. Spawns a Claude agent with web tools to find new sources, drops them into raw/, runs the compile pipeline, and emits autoresearch traces for the autoflow ratchet.
tmux-agent-launcher
Launch Claude, OpenCode, or Codex in a named tmux session from the command line or agent workflows.
test-quality-validator
Validate generated or hand-written tests for coverage completeness, correctness, and false-green risks using the canonical QA runtime, regression artifacts, and optional delivery-profile rules.
tech-spec
Generate production-ready technical specifications from product specs.
tech-spec-review
Seven-lens engineering review for technical_spec.md files, including an explicit simplicity and architectural fitness gate.
spec-to-regression
Generate browser and API regression scenarios from approved product and technical specs using the Harnessy delivery profile and regression artifact contract.