skill-promote

name	skill-promote
description	Promote trace-driven improvements from the installed skill copy back to the shared Harnessy source.
disable-model-invocation	true
allowed-tools	Read, Write, Edit, Bash, Grep, Glob
argument-hint	<skill-name> [--dry-run] [--all]

Skill Promote — Sync Improvements to Shared Source

Purpose

After /skill-improve applies changes to the installed copy at ~/.agents/skills/<skill-name>/, those improvements only exist locally. This skill promotes them back to the shared Harnessy source at tools/flow-install/skills/<skill-name>/ so they become permanent and distributable.

Inputs

• skill-name — the skill to promote (or --all to check all skills)
• --dry-run — show what would change without applying
• --all — scan all shared Harnessy skills for unpromoted improvements

Template paths are resolved from ${AGENTS_SKILLS_ROOT}/skill-promote/.

Steps

1. Detect unpromoted improvements

Compare the installed and source versions:

python3 "${AGENTS_SKILLS_ROOT}/_shared/promote_check.py" check \
    --skill "<skill-name>" \
    --installed-root "${AGENTS_SKILLS_ROOT}" \
    --source-root "<flow-repo>/tools/flow-install/skills"

The script compares manifest.yaml versions at both paths. If installed version > source version, improvements exist that haven't been promoted.

Also read ~/.agents/traces/<skill-name>/improvements.ndjson to identify which specific improvements are pending.

If no unpromoted improvements exist, report "No unpromoted improvements for <skill-name>." and stop.

2. Resolve Harnessy repo path

Find the Harnessy repo root. Check in order:

1. Current working directory — if it contains tools/flow-install/skills/, use it
2. $FLOW_REPO_ROOT environment variable
3. $HOME/.cache/harnessy/ (cached installer source)

If the Harnessy repo can't be found, report the error and ask the user to cd into it or set FLOW_REPO_ROOT.

3. Diff installed vs source

For the skill being promoted, diff all files:

diff -rq "${AGENTS_SKILLS_ROOT}/<skill-name>/" "<flow-repo>/tools/flow-install/skills/<skill-name>/"

Present the diff to the user with context from improvements.ndjson:

Unpromoted improvements for issue-flow (installed: 0.8.3, source: 0.8.0):

  imp_20260410_001 (2026-04-10): Added mobile/responsive requirement to PRD phase
    Evidence: 3 traces, pattern "mobile experience" in 2/3
    File: commands/issue-flow.md, section: Phase 2 — PRD

  imp_20260412_001 (2026-04-12): Enforce measurable acceptance criteria
    Evidence: 2 traces, category UNCLEAR_CRITERIA
    File: commands/issue-flow.md, section: Phase 2 — PRD

Files changed:
  M commands/issue-flow.md
  M manifest.yaml

If --dry-run, stop here.

4. Confirm with user

Ask: "Apply these improvements to the shared Harnessy source? (yes / no)"

5. Apply the promotion

For each changed file, copy from installed to source:

cp "${AGENTS_SKILLS_ROOT}/<skill-name>/<file>" "<flow-repo>/tools/flow-install/skills/<skill-name>/<file>"

6. Stage and show the diff

Stage the changes in the Harnessy repo:

git add "tools/flow-install/skills/<skill-name>/"
git diff --cached --stat

Show the user the staged diff for final review.

7. Suggest commit and sync

Do NOT commit automatically. Instead, suggest:

Changes staged in the Harnessy repo. Recommended next steps:

1. Review: git diff --cached
2. Commit:
   git commit -m "improve(<skill-name>): promote trace-driven improvements v<old> → v<new>

   Improvements promoted from installed copy based on decision trace evidence:
   - <imp_id>: <summary>
   - <imp_id>: <summary>

   Trace evidence: <N> traces across <M> runs, patterns: <list>"

3. Push and sync:
   - Push to remote branch
   - Use /context-sync or manual PR for review

8. Mark improvements as promoted

After the user confirms the commit happened, update ~/.agents/traces/<skill-name>/improvements.ndjson by appending a promotion record:

{
  "type": "promotion",
  "timestamp": "<ISO 8601>",
  "skill": "<skill-name>",
  "improvements_promoted": ["imp_20260410_001", "imp_20260412_001"],
  "promoted_to_version": "0.8.3",
  "commit_branch": "<branch>"
}

--all mode

When invoked with --all:

1. Scan every skill directory under ${AGENTS_SKILLS_ROOT}/ that also exists under <flow-repo>/tools/flow-install/skills/
2. Compare versions
3. Report a summary table:

Skill               Installed  Source  Unpromoted
issue-flow          0.8.3      0.8.0  3 improvements
code-review         0.2.1      0.2.1  —
prd                 0.3.0      0.2.0  1 improvement

4. Ask which skills to promote (all / specific names / none)

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-promote" --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-promote" \
    --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-promote run? (skip to finish)" If provided, capture via trace_capture.py with gate "run_retrospective" and gate-type "retrospective".

Output

• Diff summary with improvement evidence
• Staged changes in Harnessy repo (not committed)
• Suggested commit message with trace evidence
• Promotion record in improvements.ndjson