skill-reflect — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited skill-reflect (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
You run immediately after a skill session ends. Your job is to mine the conversation that just happened, extract the friction points, and write a structured log to disk so the evolver can read it later. The session is still in context — use it.
The user invokes this as /skill-reflect <skill-name> (e.g., /skill-reflect qa). If they didn't provide a skill name, ask: "Which skill should I write a session log for?"
Once you have the name, find its owning plugin. Try scanning the working directory first (works if you're in the skills repo):
ls -d plugins/*/skills/<skill-name>/ 2>/dev/nullplugins/<plugin>/skills/...).<skill-name> in both <plugin-a> and <plugin-b> — which one? (e.g., <plugin-a>/skill-name)"git branch --show-current. If plugins/ exists but the scan found nothing, the skill's plugin may live on an unmerged branch — switch to the correct branch and re-scan. Only if the branch check doesn't resolve it, ask: "Which plugin does <skill-name> belong to? (delivery, writing, track, or meta)" and use their answer as the plugin name.Review the current conversation — the skill run is still in context. Work through each of the four questions below. Be specific: quote or paraphrase actual moments from the conversation rather than speaking in generalities. A vague log is useless to the evolver.
Look for steps you added on the fly that weren't instructed by the SKILL.md — decisions you made without a written rule to follow, phases you invented, tools you reached for that the skill didn't mention. Improvisation is a signal the SKILL.md has a gap.
Hint: scan for moments where you reasoned "I'll also…", "it makes sense to…", or added a step that wasn't in a numbered phase.
Look for explicit corrections, redirections, and pushback: "no", "wait", "actually", "that's not right", "don't do that", or any message where the user steered you away from what you were doing. Each correction is a place where the skill's instructions were unclear, missing, or wrong.
Look for situations you had to navigate that the SKILL.md didn't prepare you for — ambiguous inputs, missing context, unusual project state, dependencies the skill assumed existed but didn't, questions you had to ask that a better-written skill would have answered upfront.
Based on the above, propose concrete improvements to the SKILL.md. Be specific about what would change: a new step in a phase, a clarification to an existing instruction, a fallback for an edge case, a new question to ask upfront. Vague suggestions ("improve the instructions") are not useful — name the section and describe the change.
Get the current timestamp:
date -u +"%Y-%m-%dT%H-%M"Create the directory and write the log:
mkdir -p ~/.claude/skill-sessions/<plugin>/sessions/<skill-name>Write to ~/.claude/skill-sessions/<plugin>/sessions/<skill-name>/<timestamp>.md using this exact structure:
---
skill: <skill-name>
plugin: <plugin>
date: <YYYY-MM-DD>
session_summary: <one sentence describing what was done in this session>
---
## What I had to improvise
<your findings from 2.1, or "Nothing notable." if clean>
## User corrections
<your findings from 2.2, or "No corrections." if clean>
## Edge cases not anticipated
<your findings from 2.3, or "None encountered." if clean>
## Proposed improvements
<your findings from 2.4, or "No changes recommended." if clean>If all four sections are empty (a clean run with no friction), still write the log — a record of clean sessions is signal too.
Tell the user:
Example:
Session log written to ~/.claude/skill-sessions/delivery/sessions/qa/2026-06-04T14-32.md. Top friction: the skill didn't handle the case where no Linear issue could be inferred from the branch name — Claude had to ask, and a fallback prompt should be in Phase 0.~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.