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 agentmods add skills/stainless-code/codemap/pr-comment-fact-checknpx skills add stainless-code/codemap --skill pr-comment-fact-checkgit clone --depth 1 https://github.com/stainless-code/codemapWhat 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 | $0.00098 | $0.01659 |
| Opus 5 | $0.00049 | $0.00830 |
| Sonnet 5 | $0.00020 | $0.00332 |
| Haiku 4.5 | $0.00010 | $0.00166 |
Grade A, and why
pr-comment-fact-check 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 2d 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 — 58 lines — stays where its author put it; the contents beside it link to each section on GitHub.
PR comment fact-check
When a PR has reviewer comments (human or bot), don't apply suggestions reflexively. Each comment is a claim about the code that may be wrong, partial, outdated, or stylistic.
The agent's job is to verify before acting. Auto-applying suggestions from LLM reviewers (Bugbot, Copilot, Cursor's own bot, CodeRabbit) is the most common silent regression source — they confidently propose changes that contradict project conventions because they don't have the project context the human team has codified in skills/rules.
Workflow: WORKFLOW.md.
Non-negotiables
- Never auto-apply suggestions — every comment is a claim that must be verified against the actual code (
Read), structural facts (codemap), authoritative project conventions (.agents/rules+ skills), and toolchain signals (bun run typecheck,bun run check). - Never reply "fixed!" without verifying — that's how subtle bugs land. Apply, run the project's checks, then reply.
- Never dismiss a comment without evidence — push-back is fine; evidence-free push-back wastes the reviewer's time.
- Categorize before acting — sort into ✅ correct / ❌ hallucinated / ⚠️ partial / 🕒 outdated / 💭 style. Action defaults per category below.
| Verdict | What it means | Default action |
|---|---|---|
| ✅ Correct | The code does/says what the reviewer claims; their suggestion improves it. | Apply + resolve thread. |
| ⚠️ Partially correct | Premise right, conclusion wrong (or vice versa). | Apply salvageable part + reply explaining nuance; resolve only if reviewer agrees. |
| ❌ Incorrect (hallucination) | The code doesn't do what the reviewer claims, OR the reviewer cites a "best practice" that contradicts our actual rules / TS + library versions. | Push back with evidence; leave unresolved (merge-gate exception in WORKFLOW). |
| 🕒 Outdated | The code already changed since the comment was posted. | Point at fix commit + resolve. |
| 💭 Style preference | Not enforced by any lint rule or skill; subjective. | Apply if cheap + resolve; otherwise defer to author. |
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 2d ago First seen · 58 lines · 98 tokens per session scan A 06e56f200433
pr-comment-fact-check is a skill published in the GitHub repository stainless-code/codemap (8 stars, last pushed 7d ago), licensed MIT. It adds 98 tokens to every session and 1,659 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.
Other skills, from other repositories
roam
Codebase comprehension via roam-code CLI. Use when exploring codebases, planning modifications, debugging failures, assessing PR risk, or checking architecture health. Triggers on: understanding project structure, pre-change safety checks, finding symbols/files, blast radius analysis, affected tests, health scoring…
analyze-usage
Use when asked to analyze codescout tool usage, check for error patterns, audit tool health, generate a usage report, spot anti-patterns, or clear usage statistics to start fresh.
m1nd-first
Use when investigating a repository, searching for implementation, reviewing changes, working from specs/docs, or preparing a risky code change in an environment where m1nd is available. This doctrine makes m1nd the first investigative layer before grep, glob, or manual file reads, except when the task is pure…
mcp-headless-test
Run live tests of the Ariadne MCP server using Claude Code in headless mode. Tests tool discovery, invocation, and filtering. Use when validating MCP server changes or checking integration with Claude Code.
mcp-analytics
Check Ariadne MCP tool usage analytics — session counts, tool call stats, durations, and error rates.
release
Prepare a new release of Ariadne packages. Creates changesets, verifies build/tests, and drafts the PR to main.