Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/melodic-software/claude-code-plugins/audit-noisenpx skills add melodic-software/claude-code-plugins --skill audit-noisegit clone --depth 1 https://github.com/melodic-software/claude-code-pluginsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/melodic-software/claude-code-plugins/audit-noise)<a href="https://agentmods.dev/skills/melodic-software/claude-code-plugins/audit-noise"><img src="https://agentmods.dev/badge/skills/melodic-software/claude-code-plugins/audit-noise.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00224 | $0.05986 |
| Opus 5 | $0.00112 | $0.02993 |
| Sonnet 5 | $0.00045 | $0.01197 |
| Haiku 4.5 | $0.00022 | $0.00599 |
Grade D, and why
audit-noise scanned grade D with 2 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Hidden instructionshighPrompt injection
Directives inside HTML comments, invisible characters or bidirectional overrides are read by the model and not by the person reviewing the file.
- **Opt-out markers respected.** A well-formed HTML comment line `<!-- markdown-discipline-ignore -->` covers the next paragraph, through the next blank line or heading. `<!-- markdown-discipline-ignore-line -->` covers Recursive force deletemediumDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
`token`, `password`, `api key`, `force-push`, `--force`, `rm -rf`, `destructive`, `irreversible`, Downgraded: this mod is about security review, or the phrase is quoted, so it is likely naming the pattern rather than instructing it.
How it starts
The opening of the file, as written. The whole thing — 217 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Pre-computed context
Current branch: !git branch --show-current 2>/dev/null || echo "unknown"
Uncommitted .md files (empty = none matched or the probe returned nothing): !git status --porcelain >/dev/null 2>&1 && { git status --porcelain 2>/dev/null | grep -E '\.md"?$' | head -10; :; } || echo "(git status unavailable)"
Noise findings (sample): !${CLAUDE_SKILL_DIR}/scripts/detect.sh 2>/dev/null | grep -E '^(Summary total:|Finding shape:)' | head -20 || echo "none"
Purpose
Tracked markdown accumulates nine NOISE shapes distinct from FLAVOR, owned by the sibling /docs-hygiene:compress. The surfaces include rules, skill bodies, instruction files (CLAUDE.md, AGENTS.md), docs/, and READMEs. Each shape carries a maintenance tax plus a reader-facing tax that compounds across the corpus. This skill is a read-only classifier: it surfaces candidates with treatment guidance; the author hand-applies every edit.
Three of the nine, plan-reference, conversational-antecedent, and ticket-pr-residue, carry the same names the code-side sibling /code-tidying:audit-comment-residue uses, because they are the same authoring failure landing in a different file type. Ownership splits by file type, not by shape: markdown is this skill's, everything else is the sibling's, and neither scans the other's files. The patterns are not shared code. The sibling classifies only the extracted comment portion of a line; this skill classifies whole prose, where the same words are load-bearing far more often, so its patterns are tightened accordingly and several of the sibling's cues are deliberately not carried over.
Existence pre-check (before in-page noise)
Before classifying in-page noise, ask the whole-page admission question first: could a reader with repository search derive this page's content from the code itself? A page failing admission is a deletion candidate. Its finding recommends relocate-then-delete (salvage anything admissible first), never a line-level noise treatment, and never auto-delete (this skill stays read-only).
What ships with it
13 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- context/persist-findings.md 7.9 KB
- evals/evals.json 16 KB
- evals/fixtures/legit-optouts.md 2.4 KB
- evals/fixtures/negation-shapes.md 813 B
- evals/fixtures/negation-trigger-fence.md 612 B
- evals/fixtures/noisy-rule-snippet.md 901 B
- evals/fixtures/recall-paraphrases.md 450 B
- scripts/detect.sh 15 KB runs code
- scripts/detect.test.sh 67 KB runs code
- scripts/emit-findings.sh 19 KB runs code
- scripts/emit-findings.test.sh 10 KB runs code
- scripts/lib/noise-shapes.sh 37 KB runs code
- scripts/lib/parse-concern-value.sh 6.3 KB runs code
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 217 lines · 224 tokens per session scan D 1659b110bf4f
audit-noise is a skill published in the GitHub repository melodic-software/claude-code-plugins (15 stars, last pushed today), licensed MIT. It adds 224 tokens to every session and 5,986 once invoked, about $0.0011 per session on Opus 5. A static security scan graded it D with 2 findings (hidden instructions, recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.
Other skills, from other repositories
parallel-orchestrator
Manage parallel Claude Code workstreams using git worktrees. Use when: splitting large tasks across multiple workers, coordinating parallel development, monitoring worker progress, integrating completed work, analyzing work item documents (code reviews, issue lists). Triggers: parallel, orchestrator, worktrees…
parallel-worker
Execute focused implementation tasks in a parallel workflow. Use when: working on assigned files in a worktree, making checkpoint commits, signaling dependencies or blockers, completing orchestrator-assigned tasks. Triggers: worker, checkpoint, worktree, assigned scope, commit prefix, parallel task.
build-priority-queue
For ordered processing: A search, Dijkstra, event simulation, task scheduling. Efficient min/max extraction with heap-based queue.
catch-expected-errors
For iteration with errors: catch exceptions during exploration, skip invalid cases, continue to next attempt.
compose-small-helpers
For complex behavior: build from tiny functions, chain transformations, make code read like a pipeline of operations.
count-combinations
For probability and counting: permutations, combinations, sample spaces, Monte Carlo simulation, brute-force enumeration, card/dice problems.