skill-feedback - SKILL.md Agent Skill

name: skill-feedback description: Capture ad-hoc feedback about any skill for the decision trace system. disable-model-invocation: true allowed-tools: Read, Write, Bash argument-hint: " """

Skill Feedback — Ad-Hoc Trace Capture

Purpose

Record feedback about any skill's behavior as a decision trace. This feeds the short loop (skill reads its own traces on future runs) and the long loop (/skill-improve analyzes accumulated traces to propose skill mutations).

Use this when you have feedback that doesn't correspond to a specific gate interaction — general impressions, workflow complaints, meta-observations, or suggestions.

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

Follow the Harnessy policy in .jarvis/context/docs/standards/skill-feedback-protocol.md when deciding whether feedback must be captured and which skill should receive the trace. The short rule is: capture reusable skill lessons, not empty retrospectives.

Inputs

skill-name — the skill to attach feedback to
feedback text — free-text description of the issue, suggestion, or observation

Steps

Parse arguments: extract skill name and feedback text from $ARGUMENTS.
Validate skill exists: check that ${AGENTS_SKILLS_ROOT}/<skill-name>/ or ~/.agents/skills/<skill-name>/ exists. If not, report the error and list similar skill names.
- Attach the trace to the skill that should change. Do not attach routine feedback to skill-feedback unless the recorder itself failed.

Capture the trace:

python3 "${AGENTS_SKILLS_ROOT}/_shared/trace_capture.py" capture \
    --skill "<skill-name>" \
    --gate "ad_hoc" \
    --gate-type "retrospective" \
    --outcome "approved" \
    --feedback "<feedback text>"

Confirm: report that the feedback was recorded and the trace file location.
Suggest: if the skill has 5+ traces with refinement loops, suggest running /skill-improve <skill-name>.

Output

Confirmation message with trace ID
Trace file path
Optional improvement suggestion