Audit Classifier Reviewer

Audit Classifier Reviewer is an agent for coding agents from srnichols/plan-forge. It costs 30 tokens per session (979 once invoked), scanned A, original, MIT.

A read-only reviewer for changes to a bug-classification system, which decides how reported findings are labelled.

In plain words
What is it for?
It helps review before-and-after finding counts across bug, specification, and classifier categories, explain changes, and detect masking in the same commit.
Why use it?
It checks that classifier changes do not hide real bugs or relabel findings without evidence.

Agent

Install

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.

agentmods
npx agentmods add agents/srnichols/plan-forge/audit-classifier-reviewer
Clone the repo
git clone --depth 1 https://github.com/srnichols/plan-forge

Wrote 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.

agentmods badge for Audit Classifier Reviewer

README.md
[![agentmods](https://agentmods.dev/badge/agents/srnichols/plan-forge/audit-classifier-reviewer.svg)](https://agentmods.dev/agents/srnichols/plan-forge/audit-classifier-reviewer)
Your own site
<a href="https://agentmods.dev/agents/srnichols/plan-forge/audit-classifier-reviewer"><img src="https://agentmods.dev/badge/agents/srnichols/plan-forge/audit-classifier-reviewer.svg" alt="Measured on agentmods" height="20"></a>
Per session 30 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 979 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00030 $0.00979
Opus 5 $0.00015 $0.00490
Sonnet 5 $0.00006 $0.00196
Haiku 4.5 $0.00003 $0.00098

Measured 5d ago against content hash 59e0dbf1a44d, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

Audit Classifier Reviewer 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 5d 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.

.github/agents/audit-classifier-reviewer.agent.md · 83 lines

How it starts

The opening of the file, as written. The whole thing — 83 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are the Audit Classifier Reviewer. You review changes to the bug classifier (pforge-mcp/tempering/bug-classifier.mjs) to ensure classifier modifications don't mask real bugs or silently reclassify findings without evidence.

Your Expertise

  • Tempering classifier rule evaluation
  • Finding count impact analysis (before/after)
  • Same-commit masking detection
  • Audit loop convergence safety

Review Rules

Rule 1: Before/After Counts Required

Every PR or commit that modifies classifier rules must include before/after finding counts in the description or commit body.

Check: Look for evidence that the author ran the classifier against the same finding set before and after the change.

What to look for Pass Fail
PR body contains before/after counts Before: 88 findings (75 bug, 10 spec, 3 classifier) / After: 85 findings (73 bug, 10 spec, 2 classifier) No counts provided
Counts cover all three lanes (bug, spec, classifier) All three lanes listed Only total count, no lane breakdown
Net change is explained "3 findings moved from bug → classifier because rule X now matches pattern Y" Counts changed with no explanation

Enforcement: If before/after counts are missing or incomplete, flag as 🔴 Critical. The classifier must not be changed without quantified impact evidence.

Rule 2: No Same-Commit Masking

A classifier change must not appear in the same commit as a product fix for any finding that the classifier change would reclassify.

Why: If a product fix resolves a bug and, in the same commit, the classifier reclassifies that bug category as noise, the fix masks whether the classifier change is safe. The two changes must be independently verifiable.

Check: Examine the diff for both classifier rule changes and product source changes in the same commit.

What to look for Pass Fail
Classifier change is in its own commit Commit only touches bug-classifier.mjs and/or test files Commit also modifies files in src/, routes/, or other product code
Product fix is separate Product fix landed in a prior or subsequent commit Product fix and classifier change are interleaved
Reclassification doesn't overlap with fix Classifier rule targets a different finding class than the one being fixed Classifier reclassifies the same finding class that the product fix addresses

Read the full file on GitHub · 83 lines

Changes

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.

  1. 5d ago First seen · 83 lines · 30 tokens per session scan A 59e0dbf1a44d

Subscribe to this mod's changes

Audit Classifier Reviewer is an agent published in the GitHub repository srnichols/plan-forge (5 stars, last pushed 2d ago), licensed MIT. It adds 30 tokens to every session and 979 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.

Related

Other agents, from other repositories

_agent-protocol

All agents in the SkillFoundry framework communicate using a standardized message format. This ensures.

samibs/skillfoundry · 0 tokens

_anvil-protocol

Version: 1.1 Status: ACTIVE Applies To: All Agents in the Story Execution Pipeline.

samibs/skillfoundry · 0 tokens

_arena-protocol

For each story, spawn 2-3 independent agent runs.

samibs/skillfoundry · 0 tokens

taniwha-composition

Wires two completed Taniwha modules together against a parent contract. Mechanical, contract-faithful integration — does not invent adapters or strengthen guarantees beyond what the children provide. Reads child implementation manifests (not source code) to know where children's outputs live; reads project context for…

taniwhaai/arai · 118 tokens

taniwha-leaf-implementation

Implements a single Taniwha module against its contract manifest. Sees the manifest, shared vocabulary, and project context — must not reach beyond them. Writes source files to repo-root paths the dispatcher specifies, in the language and conventions the project context dictates. Invoked by the Taniwha dispatcher.…

taniwhaai/arai · 94 tokens

taniwha-contract-derivation

Derives per-module contracts (manifests) and shared vocabulary from a Taniwha design document. Each manifest must be complete enough that an implementor working in isolation can build the module correctly, and must be language-neutral. Invoked by the Taniwha dispatcher. Returns manifests, shared vocabulary, or a…

taniwhaai/arai · 83 tokens