rfs-proposal
File a Request for Skill (RFS) when a skill gap is detected during execution.
- License 许可
- MIT
- 最近更新
- 2026-03-11
- Stars
- ★ 0
- Forks
- 1
01 前言 Front Matter
- name
- rfs-proposal
- description
- File a Request for Skill (RFS) when a skill gap is detected during execution.
- disable-model-invocation
- true
03 SKILL.md 正文
Request for Skill (RFS) Proposal
Use this skill when you detect a skill gap — a task type or behavioral pattern that no existing skill covers adequately.
Golden rule: Never block on a gap. Complete the task first, then file the RFS at the end.
Rate Limit
At most 1 RFS per run. If you detect multiple gaps, file the most impactful one and note the others in the issue body.
Step 1: Check for Duplicates
Before filing, search for existing RFS issues:
gh issue list --label "rfs" --state open --json number,title --jq '.[] | "#\(.number) \(.title)"'
If a matching RFS already exists, skip filing. Optionally add a comment noting the new occurrence.
Step 2: Create the RFS Issue
File a GitHub Issue using the structured template:
gh issue create \
--label "rfs,rfs:proposed" \
--title "RFS: <proposed-skill-name>" \
--body "$(cat <<'BODY'
## Request for Skill
| Field | Value |
|-------|-------|
| **Skill name** | `<kebab-case-name>` |
| **Gap type** | `new-skill` or `skill-extension` |
| **Extends skill** | `<existing-skill-name>` (if skill-extension, otherwise N/A) |
| **Installation target** | `consumer` (.po-agent/skills/) or `framework` (built-in) |
### Trigger Context
What the agent was doing when the gap was detected:
- Task type: <description>
- Skills consulted: <list of /skill-name checked>
- Why existing skills fell short: <explanation>
### Desired Behavior
What the skill should teach the agent to do:
<detailed description of the behavioral pattern>
### Draft SKILL.md
<If you have enough context, draft the SKILL.md content here. Otherwise write "Needs human authoring.">
```markdown
---
name: <skill-name>
description: <one-line description>
disable-model-invocation: true
---
# <Skill Title>
<Instructions>
Additional Gaps (if any)
<List other gaps detected in this run that were NOT filed as separate RFS issues>
Source
- Run: <link to the GitHub Actions run or PR, if available> BODY )"
Adapt the template fields based on the actual gap detected. The key fields are:
- **Skill name**: kebab-case, descriptive
- **Gap type**: `new-skill` if nothing covers this area; `skill-extension` if an existing skill is close but lacks guidance
- **Installation target**: `consumer` for project-specific skills, `framework` for skills that benefit all PO Agent users
## Step 3: Notify on Current Thread
After filing the RFS, add a short note to the current PR or thread:
> ✅ Task completed. Filed [RFS #N](link) for a missing capability: **<skill-name>** — <one-sentence description of the gap>.
This goes in your normal response — do NOT create a separate comment just for the notification.
## Step 4: Log to RFS Log
Append a row to `.po-agent/rfs-log.md` (create the file if it doesn't exist):
```bash
# Create the log file with header if it doesn't exist
if [ ! -f .po-agent/rfs-log.md ]; then
mkdir -p .po-agent
cat > .po-agent/rfs-log.md << 'HEADER'
# RFS Log
| Date | RFS # | Skill Name | Gap Type | Status |
|------|-------|------------|----------|--------|
HEADER
fi
# Append the new entry
echo "| $(date +%Y-%m-%d) | #<NUMBER> | \`<skill-name>\` | <gap-type> | proposed |" >> .po-agent/rfs-log.md
Commit and push the log update:
git add .po-agent/rfs-log.md && git commit -m "chore: log RFS #<NUMBER>" && git push
04 同仓库更多 Skills
story-workflow
9-phase workflow for implementing user stories. Covers requirements through verification.
review-handling
Process review comments from AI review or human reviewers. Evaluate, respond, and resolve or escalate.
pr-creation
Create GitHub pull requests with proper format.
playwright-verify
Browser verification using Playwright MCP. Navigate, interact, screenshot.
merge-conflicts
Resolve merge conflicts by rebasing. Use when PR mergeable status is CONFLICTING.
mandatory-gates
Pre-implementation gates. MUST be followed before writing any code. Investigation, confidence assessment, and scope confirmation.