CategoryWeightCategory scoreContribution

Securityprompt, exec, net, exfil, eval

35%

30.8 pts

Supply chainhash, typosquat, maintainer, lockfile

20%

100

20.0 pts

Maintenancestaleness, pinning, CI

15%

100

15.0 pts

TransparencySKILL.md, perms, README

15%

100

15.0 pts

Communityinstalls, verify, response

15%

100

15.0 pts

Findings & checks · 1 flagged

Securityscore 88 · 1 finding

MEDIUMInstruction telling the agent not to ask for approvalSS-SKILL-INJECT-DONT-ASK-01 · Prompt injection · CONTRIBUTING.md

MEDIUM — it fires on intent; the real damage depends on the host agent's own approval-gating.

Why it matters

The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.

The exact value spotted

excerptCONTRIBUTING.md· markdown

272. **Idempotent.** Re-running a bootstrap produces the same state. Never overwrite user cont

… (56 chars elided on L27)

283. **Profiles are data, never code.** No `eval`, no embedded scripts, no remote includes.

294. **`gh` integration is always best-effort.** Skip with a logged reason; never fatal, never

… (18 chars elided on L29)

305. **MCP boundary.** Shell scripts never invoke MCP tools — only the skill layer does.

316. **No internal references in code.** Names must be self-descriptive. Traceability belongs

… (15 chars elided on L31)

Occurrences

1 occurrence · at L29

How to fix

Remove the approval-skipping instruction, or scope it narrowly to a specific safe, reversible action.

Delete blanket "don't ask / no need to confirm" directives from the skill.
If the skill is a genuine autonomous job, restrict the opt-out to a named non-destructive action rather than all actions.

Framework references

OWASPLLM01 ATLASAML.T0051

Trace & refs

ruleSS-SKILL-INJECT-DONT-ASK-01sha256183eb789192b5159…rubric 365aacaView on GitHub

Supply chainscore 100 · 0 findings

✓—All supply chain checks passedNo findings in this category for the latest scan.pass

Maintenancescore 100 · 0 findings

✓—All maintenance checks passedNo findings in this category for the latest scan.pass

Transparencyscore 100 · 0 findings

✓—All transparency checks passedNo findings in this category for the latest scan.pass

Communityscore 100 · 0 findings

✓—All community checks passedNo findings in this category for the latest scan.pass

The primary manifest — the file an agent reads to learn what this artifact does.

skills/optimize-claudemd/SKILL.md · 1 file

skills/optimize-claudemd/SKILL.md3.4 KB · Markdown

name: optimize-claudemd

description: "Optimize CLAUDE.md based on actual usage patterns. TRIGGER ON: 'optimize CLAUDE.md', 'improve CLAUDE.md', 'make CLAUDE.md better', 'trim CLAUDE.md', 'what is unused in CLAUDE.md', 'CLAUDE.md is too big', 'shrink CLAUDE.md', 'which docs are never read', 'CLAUDE.md usage', 'optimize context'. Analyzes which sections, docs, and commands Claude actually uses and recommends removals, compressions, and additions."

optimize-claudemd — Optimize CLAUDE.md Based on Usage

You are the optimize-claudemd skill. You analyze how Claude actually uses the CLAUDE.md file — which sections are referenced, which docs are read, which commands are run — and recommend optimizations to maximize the value of every byte.

When to trigger

User asks to optimize, improve, trim, or shrink CLAUDE.md
User asks what's unused or which docs are never read
User says CLAUDE.md is too big or wants to improve context quality
After significant project evolution (new features, removed docs)
Health score shows CLAUDE.md-related deductions

DO NOT trigger on: general CLAUDE.md questions (use explain-state), initial CLAUDE.md generation (use bootstrap), or manual CLAUDE.md editing.

Prerequisites

Usage tracking must be enabled (memory/claudemd-usage.json must exist)
At least 10 sessions of tracked data (override with --force)
CLAUDE.md must have nyann markers ( / )

If tracking is not enabled, offer to enable it: create memory/claudemd-usage.json with the empty template, then explain that data will accumulate over future sessions.

Execution flow

Phase 1: Analyze usage

Run bin/analyze-claudemd-usage.sh --target . to get the usage analysis.
Check sufficient_data — if false, explain how many more sessions are needed and offer --force.

Phase 2: Present findings

Show the analysis summary:

Total sessions tracked
Section-by-section density table (section, bytes, references, density, verdict)
Unused docs (referenced in CLAUDE.md but never read by Claude)
Missing commands (frequently run but not documented)
Budget usage (bytes used / remaining)

For each recommendation, explain:

What to change (remove, compress, or add)
Why (density score, usage count)
Estimated byte savings

Phase 3: Apply optimizations

Ask which recommendations to apply (all, specific ones, or none).
Run bin/optimize-claudemd.sh --target . --profile <path> --dry-run to preview.
Show the before/after diff.
On confirmation, run without --dry-run.
Report total bytes saved and new budget status.

Phase 4: Next steps

Suggest running /nyann:doctor to see the health score impact.
If tracking was just enabled, explain that optimization will improve with more usage data.

Key constraints

Only modifies content within  /  markers
Preserves all user content outside markers
Respects 3KB soft budget and 8KB hard cap
Never removes sections without showing the user what will be removed
Compression is advisory — the skill shows what could be shortened but the actual rewrite is done by the LLM, not the bash script

When to hand off

"Regenerate CLAUDE.md from scratch" → gen-claudemd skill (full

regeneration from profile + stack, not usage-driven pruning).

"Check repo health" → doctor skill.
"What does nyann see here?" → explain-state skill.

optimize-claudemd— agent skill

Is optimize-claudemd safe to install?

optimize-claudemd — Optimize CLAUDE.md Based on Usage

When to trigger

Prerequisites

Execution flow

Phase 1: Analyze usage

Phase 2: Present findings

Phase 3: Apply optimizations

Phase 4: Next steps

Key constraints

When to hand off

Audit the pieces. Scan the whole. Decide.