Install
$ agentstack add skill-mouchegmouradian-claude-code-skills-skill-improvement ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
Skill Improvement
This skill captures knowledge gained during a working session and feeds it back into the relevant skill — keeping skills accurate and up to date without manual maintenance.
When to trigger
Trigger this skill whenever, during a session, you observe any of the following:
- A skill's instructions were incomplete or wrong and you had to correct course
- You discovered an edge case the skill didn't cover
- The user corrected you in a way that implies a skill gap
- A fix, workaround, or pattern emerged that should be documented for next time
- The user explicitly says "update the skill", "add this to the skill", etc.
- At natural session wrap-up points when a skill was actively used — proactively check if anything
is worth capturing, even if the user didn't mention it
Don't wait to be asked. If you notice something, surface it.
Skill repo location
The skills git repo is at: ~/Projects/claude-code-skills/ Skills source lives at: ~/Projects/claude-code-skills/skills/ The ~/.claude/skills/ directory contains symlinks into the above.
The improvement flow
Step 1 — Identify and summarize the gap
When you detect an improvement opportunity, pause and tell the user:
> "I noticed something worth capturing in the [skill-name] skill: [one-sentence description of the gap or fix]. > Want me to propose a patch?"
Keep it brief. Don't write the patch yet.
Step 2 — Draft the patch
If the user says yes (or "go ahead", "sure", etc.), draft the minimal change to SKILL.md:
- Show a unified diff or clearly marked before/after block
- Keep changes surgical — fix only what was wrong, don't refactor unrelated sections
- If adding new content, place it in the most logical existing section rather than appending at the end
- If the change is substantial, briefly explain your reasoning
Then ask:
> "Does this look right, or would you like to adjust anything before I commit it?"
Step 3 — Apply and commit (only after explicit approval)
Once the user approves (explicit "yes", "looks good", "commit it", etc.):
cd ~/Projects/claude-code-skills
# Create a branch
git checkout -b skill//
# Apply the change to the actual skill file
# (edit ~/Projects/claude-code-skills/skills//SKILL.md)
# Stage and commit
git add skills//SKILL.md
git commit -m "skill():
Context: "
# Report back
git log -1 --oneline
Branch naming convention: skill// Examples:
skill/zetic-melange/fix-npu-init-orderskill/pdf-reading/add-scanned-doc-handlingskill/skill-improvement/clarify-trigger-conditions
Commit message format:
skill():
Context:
Step 4 — Tell the user what to do next
After committing, tell the user:
> "Branch skill// is ready. Review with git diff main skill// in > ~/Projects/claude-code-skills, then merge when you're happy with it."
What makes a good skill patch
Do capture:
- Missing steps that caused errors or confusion
- Edge cases that required manual correction
- Clarifications to ambiguous instructions
- New patterns or conventions that emerged from real use
- Prerequisites or environment requirements that were assumed but not stated
Don't capture:
- One-off project-specific details that don't generalize
- Fixes to the user's code (not the skill's instructions)
- Speculative improvements not grounded in something that actually went wrong
- Style rewrites that don't change meaning
Multiple improvements in one session
If several gaps were found, handle them one at a time — one branch per logical change. Don't batch unrelated fixes into one commit.
If the affected skill is unclear
Sometimes a gap spans multiple skills or belongs in a new skill. In that case:
- If it fits mostly in one skill, put it there and note the overlap
- If it clearly belongs in a new skill, say so and offer to kick off the skill-creator flow instead
- If unsure, ask the user before proceeding
Relation to skill-creator
This skill is about targeted patches to existing skills based on real session observations. For creating a new skill from scratch, or doing a major rewrite with evals, use the skill-creator skill instead. The two can hand off to each other: if a patch grows large enough to warrant a full eval cycle, say so and suggest switching to skill-creator.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: mouchegmouradian
- Source: mouchegmouradian/claude-code-skills
- License: MIT
- Homepage: https://mouchegmouradian.github.io/claude-code-skills/
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.