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 skills add tmusser/ai-engineering-skills --skill tool-noise-guardgit clone --depth 1 https://github.com/tmusser/ai-engineering-skillsWrote 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/tmusser/ai-engineering-skills/tool-noise-guard)<a href="https://agentmods.dev/skills/tmusser/ai-engineering-skills/tool-noise-guard"><img src="https://agentmods.dev/badge/skills/tmusser/ai-engineering-skills/tool-noise-guard/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/tmusser/ai-engineering-skills/tool-noise-guard"><img src="https://agentmods.dev/badge/skills/tmusser/ai-engineering-skills/tool-noise-guard.svg" alt="Reviewed on agentmods" width="80" 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 5.1 | $0.00039 | $0.01308 |
| Opus 5 | $0.00019 | $0.00654 |
| Sonnet 5 | $0.00008 | $0.00262 |
| Haiku 4.5 | $0.00004 | $0.00131 |
Grade A, and why
tool-noise-guard scanned grade A with 0 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 11d ago.
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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 139 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Tool Noise Guard
Purpose
Reduce the forward propagation of tool-call bloat during tool-heavy work.
Compact forward, not backward. This skill cannot remove raw tool output that the runtime already injected into context. It prevents that cost from multiplying by re-quoting, re-summarizing, and re-fetching the same low-value metadata on later turns.
Never trade required evidence for token savings.
When to use
Optional passive guardrail. Once explicitly enabled for a tool-heavy task or session, stay active until the user disables it or the task no longer uses tools.
Stay silent by default. Intervene behaviorally when one or more of these signals appear:
- successive tool results repeat the same wrapper, schema, or stable metadata
- a list/search response contains many fields but only a few affect the next decision
- the same resource is fetched or polled again without a concrete freshness reason
- a follow-up call can reuse an existing ID, cursor, path, or resource reference
- verbose tool envelopes are being copied into summaries, handoffs, or user-facing prose
Do not compact aggressively when exact raw payload shape is itself the evidence, when results conflict, when an error is not yet understood, or when a mutation/audit step requires exact confirmation.
Inputs
- Current task and next decision
- Recent tool calls and results
- Resource identifiers, paths, cursors, commit SHAs, line references, and timestamps
- Current mutation, verification, pagination, or freshness state
- Any exact fields required by the next tool call
Workflow
- Before repeating a tool call, ask: what new information can this call change?
- Reuse already-known resource anchors instead of rediscovering the same object.
- For repeated result shapes, classify fields into:
anchor— needed to identify, reopen, mutate, cite, paginate, or verify the resourcedelta— changed since the previous relevant resultevidence— affects correctness, status, risk, or the next decisionnoise— repeated wrapper or metadata with no current operational use
- Carry
anchor + delta + evidenceforward. Do not restatenoise. - Prefer a targeted follow-up operation over rehydrating a full resource when both answer the same question.
- If polling mutable state, poll only when a state transition can change the next action. Do not immediately repeat an unchanged status call for reassurance.
- When several equivalent reads are needed, prefer a safe batch/list operation when the tool supports one and batching does not widen scope or hide per-item failures.
- If compacting a result cluster, retain a minimal working digest rather than another prose summary.
- Re-expand from the original tool/resource only when the next decision genuinely needs a field that was not retained.
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.
- 11d ago First seen · 139 lines · 39 tokens per session scan A 18a5eee73647
tool-noise-guard is a skill published in the GitHub repository tmusser/ai-engineering-skills (4 stars, last pushed yesterday), licensed MIT. It adds 39 tokens to every session and 1,308 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
git-ai-search
Search and restore AI conversation context from git history.
brain-page
Operating manual for reading and writing a project's brain — every read and write goes through the bundled zero-dependency brain CLI; never hand-edit brain files. Read it before creating or modifying any page or root page.
brain-setup
Bootstrap the Open Project Brain Standard into the current project — prefer brain init (ensure BRAIN.md, scaffold empty brain brainRoot-aware, default-wire CLAUDE.md + AGENTS.md). Optionally install a pre-commit hook and a Claude Code or Codex SessionStart hook.
brain-bootstrap
Seed a freshly-scaffolded brain with real project knowledge — on an existing (brownfield) project read the code, docs, and git log to draft the six root pages and capture key historical decisions; on a near-empty (greenfield) project interview the user. Every write goes through the brain CLI. Run it after brain-setup.
immune
Hybrid adaptive memory: Cheatsheet (positive patterns pre-generation) and Immune (negative patterns post-generation) with Hot/Cold tiered auto-learning. Triggers on: "scan for errors", "immune scan", "check output quality", "antibody scan". NOT for PR review (use pr-review) or repo audits (use repo-sentinel).
usage-audit
Audit a Claude Code setup for token waste and context bloat. Checks MCP servers, CLAUDE.md, skills, and settings against bloat filters. Triggers on: "audit my context", "usage audit", "token audit", "context bloat". NOT for codebase audits.