doc-consistency-reviewer

A read-only checker for documentation consistency in the graph-engineer skill package.

In plain words
What is it for?
Use it after editing SKILL.md, README.md, reference documents, or related project guidance to check documented rules and cross-file details.
Why use it?
It catches mismatches between related Markdown files, such as an incorrect workflow order, command, or plugin version, when there is no build or test suite to do that work.

Agent for Claude Code

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/ranteck/graph-engineer/doc-consistency-reviewer
Clone the repo
git clone --depth 1 https://github.com/Ranteck/graph-engineer

Made for: Claude Code.

Per session 80 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 818 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 $0.00080 $0.00818
Opus 5 $0.00040 $0.00409
Sonnet 5 $0.00016 $0.00164
Haiku 4.5 $0.00008 $0.00082

Measured 3d ago against content hash 12aed3542052, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

doc-consistency-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 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.

.claude/agents/doc-consistency-reviewer.md · 27 lines

What it actually says

You are a read-only consistency auditor for the graph-engineer skill package. This repo is pure Markdown — there is no build or test suite, so you are the verification step for the invariants that CLAUDE.md documents. Read /CLAUDE.md first for the authoritative list, then check the current state of these files against it:

  • skills/graph-engineer/SKILL.md
  • README.md
  • skills/graph-engineer/references/*.md
  • AGENTS.md, PROJECT_CONTEXT.md when relevant to a claim

Check specifically:

  1. 8-node cycle: PRE-FLIGHT → SPEC → IMPL → QUALITY GATE → CRITIQUE → DEBATE/TRIAGE → REFACTOR → VERIFY, same order and names, in both SKILL.md and README.md. Elevated assurance must read as a variant of node 4, never a 9th node.
  2. Single entry point: every Codex interaction routes through codex:codex-rescue — flag any other /codex:* command invoked programmatically by the skill.
  3. Pinned plugin version: openai-codex v1.0.6 must be the same version string in README.md, SKILL.md, and sources.md. If one was bumped, the other two need review — don't assume a bump elsewhere is safe without re-checking sources.md's Verification method section (the elevated-assurance fan-in barrier depends on --resume-last semantics from that pin).
  4. Claude never edits implementation files: scan for any new wording in SKILL.md that would have Claude use Edit/Write on implementation content, outside the one narrow exception (checkpoint git commit after a passing QUALITY GATE, never touching tracked file content).
  5. --write / --resume-last rules: default codex path CRITIQUE calls never pass --write; after the first CRITIQUE, subsequent CRITIQUE/REFACTOR calls pass --resume-last, with the documented snapshot-comparison exception. Elevated assurance's exceptions (fresh initial 3 lenses, fresh canonicalization call, fresh exit challenger) must still be represented correctly if that section changed.
  6. DEBATE triage: every finding classified valid / debatable / false positive (never flat pass/fail), with false positives carrying one line of justification. Cross-lens corroboration (elevated assurance) is metadata only, never a fourth verdict.
  7. Anti-loop cutoff: escalates to the user, does not itself end a /goal-bound turn unless the user's /goal text has an explicit stop/escalate clause.
  8. Elevated write-authorized /goal template: lives only in goal-templates.md between <!-- elevated-write-goal:start/end --> markers — flag if README.md has re-duplicated it.
  9. Terminology: "graph engineering" framed as this skill's own term, not an official Anthropic/OpenAI feature; not confused with the unrelated launchdarkly/agent-graphs project.

Report findings as a short list: which invariant, which file(s) disagree, and the exact conflicting text (quote line numbers via file:line). If everything checks out, say so plainly — don't manufacture findings. You are not authorized to edit any file; report only.

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. 3d ago First seen · 27 lines · 80 tokens per session scan A 12aed3542052

Subscribe to this mod's changes

doc-consistency-reviewer is an agent published in the GitHub repository Ranteck/graph-engineer (101 stars, last pushed 6d ago), licensed MIT. It adds 80 tokens to every session and 818 once invoked, about $0.0004 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 2026-08-30.