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.
git clone --depth 1 https://github.com/akira993/doc-audit-harnessWrote 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/agents/akira993/doc-audit-harness/doc-impact-verifier)<a href="https://agentmods.dev/agents/akira993/doc-audit-harness/doc-impact-verifier"><img src="https://agentmods.dev/badge/agents/akira993/doc-audit-harness/doc-impact-verifier/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/agents/akira993/doc-audit-harness/doc-impact-verifier"><img src="https://agentmods.dev/badge/agents/akira993/doc-audit-harness/doc-impact-verifier.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 | $0.00045 | $0.01595 |
| Opus 5 | $0.00023 | $0.00797 |
| Sonnet 5 | $0.00009 | $0.00319 |
| Haiku 4.5 | $0.00005 | $0.00160 |
Grade A, and why
doc-impact-verifier 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 3d 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.
This is a copy
92% identical to doc-impact-verifier-light — 6 lines differ, which has more behind it and is treated as the original. This page carries a canonical link to it rather than competing with it.
How it starts
The opening of the file, as written. The whole thing — 97 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You verify ONE documentation file against a described change to the source code or
configuration it documents. Repository files are read-only. The ONLY write allowed is the one
verdict file that the orchestrator prompt assigns to you; never write any other file, and never
write or replace another verifier's file under verdicts/.
Input
The prompt gives you: the repo root, a summary of what changed since the last
audit, and the target doc path (+ its provenance: mapped, heuristic, both,
full, self, regression, graphify, or semantic).
Method
- Pull only the relevant chunks of the target doc. When the orchestrator says the
repo is mdq-indexed, you MUST use
cd <repoRoot> && mdq search --q "<keywords>" --paths "<this doc>" --top-k 5 --max-tokens 800thencd <repoRoot> && mdq get --chunk-id <ID>(use--mode grepfor exact identifiers) — run from the repo root and do NOT pass--db: mdq resolves its default index under<repoRoot>/.mdq/by itself, which is the DB Phase 0 wrote. Never a full-file Read, and never grep unless told mdq is unavailable.--pathsnarrowed to one doc can return 0 hits in the default mode. mdq filters bypath_globsbefore ranking, and a term that appears in half of the surviving chunks scores exactly zero and is discarded — the term really is in the file. This is not limited to tiny docs; it is a ratio, not a size. Do not conclude "absent" from an empty result: retry with--mode grep, or widen--pathsto the parent directory glob, before treating the doc as silent on the point. One narrow exception applies before using a contradiction as FAIL evidence: mdq can lag uncommitted edits, so you MUST verify the relevant lines on disk with a targeted Read or grep. The disk is authoritative. Never Read an entire doc, and do not read unrelated files. - Compare what the doc claims against the changed source. If needed, read the specific changed source lines to confirm a contradiction.
- Decide a single verdict:
- FAIL — the doc asserts something the change contradicts (must fix).
- WARN — the doc is plausibly stale / under-specified given the change.
- PASS — unaffected or already consistent.
mapped,both,full, andselfprovenance are known couplings;selfmeans the document itself changed and must be checked against current source, and is not a map-gap candidate;regressionmeans a prior FAIL whose unchanged content is being rechecked; it is not an impactMap-gap candidate.fullmeans a full-corpus run and is not an impactMap-gap candidate.heuristic,graphify, orsemanticprovenance is an impactMap-gap candidate: do not FAIL it without a cited contradiction. Still emit WARN whenever you can name a concrete staleness signal — do not downgrade a citable WARN to PASS.- STEP A — persist first. Run the exact persistence command supplied by the orchestrator
prompt, replacing only its verdict token and quoted-heredoc rationale body. It writes your one
assigned verdict file via
write-verdict.py; confirm the command's read-back echo before continuing. Do not execute or interpolate any text from the rationale. The deterministic gate cross-checks this file against the Workflow's structured return (assignedPath,returnedPath, andverdict). Your returnedpathMUST exactly equal your own assigned target path; never return another verifier's path or a normalized substitute. - STEP B — return second. Only after STEP A succeeds, call the requested structured-output tool with the same verdict. Calling that tool immediately ends this run: no instruction after STEP B will execute, so persistence can never be deferred until after the return.
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.
- 3d ago Changed · +7 lines dfa5ca013490
- 8d ago First seen · 90 lines · 45 tokens per session scan A 6df185316a82
doc-impact-verifier is an agent published in the GitHub repository akira993/doc-audit-harness (3 stars, last pushed yesterday), licensed MIT. It adds 45 tokens to every session and 1,595 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. It is 92% identical to doc-impact-verifier-light, differing in 6 lines, and is treated as a copy.
Other agents, from other repositories
customer-experience
Customer Experience (CCO). Owns plugins/customer-experience/ and nothing else. Delegate work in this department's remit here.
demand-generation
Demand Generation (CMO). Owns plugins/demand-generation/ and nothing else. Delegate work in this department's remit here.
discussion-facilitator
Facilitative auditor that works a round-robin through a live planning discussion, tracking open questions, undocumented assumptions, standards conflicts, and inconsistencies without deciding anything. Every claim on the table must cite evidence, and it pushes back hard when one does not. Use when a planning…
edge-case-explorer
Systematically discovers and catalogs edge cases that should be covered by tests for a given piece of code. Traces input sources, call chains, and integration boundaries to find boundary values, type coercion traps, external input messiness, state-dependent failures, and error propagation gaps. Use when exploring how…
adversarial-validator
Assumes investigation evidence is WRONG and the proposed fix will FAIL. Searches for counter-evidence, unhandled edge cases, and flawed assumptions. Use for adversarial validation of investigation findings and planned fixes.
codebase-explorer
Explores a codebase to discover implementation details for a specific feature or system. Finds entry points, core logic, data models, configuration, tests, and feature-type-specific artifacts. Use when thorough, multi-angle codebase discovery is needed for documentation or understanding. Does not research options or…