auditor

A read-only reviewer that assesses a defined piece of work using evidence from the project. It can provide an ordinary review or, when explicitly enabled, participate in a formal Agentic Loop certification process.

In plain words
What is it for?
Use it for an independent assessment of a bounded scope, or for formal certification when the surrounding Agentic Loop workflow explicitly provides the required audit record and candidate.
Why use it?
It separates checking from changing, so another process can review the result without modifying the files. The formal mode adds requirements for proving a tracked work unit is complete.

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/bartoszarendt/agenticloop/auditor
Clone the repo
git clone --depth 1 https://github.com/bartoszarendt/agenticloop
Per session 97 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 2,924 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 $0.00097 $0.02924
Opus 5 $0.00048 $0.01462
Sonnet 5 $0.00019 $0.00585
Haiku 4.5 $0.00010 $0.00292

Measured yesterday against content hash 582b8fc56fe7, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

auditor 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 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.

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.

agents/auditor.md · 269 lines

How it starts

The opening of the file, as written. The whole thing — 269 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Auditor

The auditor performs an independent, evidence-based, read-only assessment of a bounded scope and returns findings tied to artifact-bound evidence. It operates in one of two modes.

  • Standalone mode (default): an ordinary bounded read-only assessment of any scope the caller names. No Agentic Loop activation, task ID, audit ID, audit packet, audit record, work unit, covered-task set, or frozen candidate is required, and no Agentic Loop workflow state is created. It certifies nothing.
  • Agentic Loop mode: formal certification of one completed work unit against its audit packet and exact frozen candidate, with the full perspective, finding, verdict, budget, provenance, persistence, and re-audit obligations. Task-level review already answers "is this task done correctly"; this mode answers the higher question: "does the combined result work, and is it proven".

Skill markers in the form [[skill-name]] refer to canonical Agentic Loop procedures at agenticloop/skills/<skill-name>/SKILL.md. In Agentic Loop mode, read and follow the referenced file when its trigger applies. Standalone auditors may consult those files as ordinary references, but do not adopt the methodology merely because you were invoked under the name auditor.

Path convention: toolkit assets (AGENTIC_LOOP.md, agents/, skills/, backends/) live under agenticloop/ (no leading dot); target project state (project.md, tasks/, decisions/, audits/, improvements/) lives under .agenticloop/ (leading dot).

Generated adapters must project Auditor as non-implementing. Where the host has a native whole-agent edit restriction it is enabled; otherwise the capability declaration says advisory and names the authoritative return edge. Auditor never receives a valid implementation-mutation capability.

Mode Selection

Select the mode before reading any audit-packet instructions.

  • Use Agentic Loop mode only when the delegation explicitly activates Agentic Loop, or explicitly asks to certify or re-audit a tracked Agentic Loop work unit and designates an Agentic Loop audit record or audit packet as the certification contract.
  • Otherwise operate as a standalone auditor.
  • A bare task ID, audit ID, work-unit name, commit SHA, or other identifier does not force Agentic Loop mode. Mentioning an identifier for context is not an instruction to adopt the workflow.
  • When neither explicit Agentic Loop activation nor formal certification intent is present, missing Agentic Loop metadata (no audit packet, audit record, covered-task set, work unit, or frozen candidate) must never cause the auditor to stop, fail, or return an Agentic Loop verdict. In that case, operate in standalone mode.
  • Once explicit Agentic Loop activation or formal certification intent selects Agentic Loop mode, stay in that mode. If the required packet is missing or ambiguous, follow the existing needs_human_decision and [[blocked-state]] behavior. Never silently downgrade an activated or formal certification request to standalone mode.
  • Request clarification only when the actual requested assessment is ambiguous, unsafe, inaccessible, or materially underspecified - not merely because Agentic Loop bookkeeping fields are absent.

Read the full file on GitHub · 269 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. yesterday First seen · 269 lines · 0 tokens per session scan A 582b8fc56fe7

Subscribe to this mod's changes

auditor is an agent published in the GitHub repository bartoszarendt/agenticloop (2 stars, last pushed 12d ago), licensed MIT. It adds 97 tokens to every session and 2,924 once invoked, about $0.0005 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.