Borrowing it
Nothing to install: this file belongs to ForgeyClap/claude-forge. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/ForgeyClap/claude-forge/main/.claude/skills/forge-graded-verify/SKILL.mdgit clone --depth 1 https://github.com/ForgeyClap/claude-forgeWrote 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/skills/forgeyclap/claude-forge/forge-graded-verify)<a href="https://agentmods.dev/skills/forgeyclap/claude-forge/forge-graded-verify"><img src="https://agentmods.dev/badge/skills/forgeyclap/claude-forge/forge-graded-verify.svg" alt="Measured on agentmods" 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 | $0.00752 |
| Opus 5 | $0.00023 | $0.00376 |
| Sonnet 5 | $0.00009 | $0.00150 |
| Haiku 4.5 | $0.00005 | $0.00075 |
Grade A, and why
forge-graded-verify 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 7d 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 — 24 lines — stays where its author put it; the contents beside it link to each section on GitHub.
forge-graded-verify — advisory rubric-scored verification (scout #7, 2026-07-13)
Fills the graded-quality axis Forge deliberately lacks: forge-verify checks structural bookkeeping ("did the agent close its tickets"); forge-evals is intentionally deterministic-binary so it can safely gate git reverts. Neither scores subjective ANSWER QUALITY. This skill adds that — advisory only. Method reimplemented from the DeepVerifier failure-taxonomy pattern (arXiv 2601.15808); no code/dataset from that repo is used.
When to use
High-stakes answer-quality work where "did it pass an assertion" isn't enough: RAG answers, research syntheses, scraped-data summaries, prediction rationales. NOT for code/build gating (that stays with forge-evals + integration-gate). Reserve for genuinely high-stakes tasks — it costs an LLM review pass; don't duplicate ultra-review.
How it works (advisory, bounded, never gates irreversible actions)
- Pick the rubric for the domain from
.claude/config/rubrics/<domain>.json. Each criterion has:id,descriptor(4-level: 1=poor … 4=excellent, what each level means), andraise(what would move the score up). - Dispatch review-boss as a GRADED verifier (not the binary reviewer): it reads the output + the source/context, and returns per-criterion
{id, score 1-4, evidence, feedback}via verification-by-decomposition (judge each criterion separately, cite evidence). - Bounded single rework loop: if any required criterion scores < the threshold (default 3), emit ONE
rework_task_created→rework_assignedto the owning Boss with the NL feedback, re-grade ONCE, respect the usage-guard. No open-ended loops. - Log advisory only: record the graded result via the existing
gate_evaluated/lead_review_completedevents (do NOT invent event types). NEVER let a graded score gate an irreversible action (deploy, git revert, send) — deterministic forge-evals / owner approval stay authoritative there.
Honesty
LLM-graded scores are judgment, not ground truth (a judge can mislabel). Frame results as advisory quality signals; cite per-criterion evidence; the reviewer model should differ from the builder model (evaluator independence). Rubrics are owner-editable JSON — start from the shipped ones and ADAPT per project, don't drop-in.
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.
- 7d ago First seen · 24 lines · 45 tokens per session scan A 80b1ce158ca6
forge-graded-verify is a skill published in the GitHub repository ForgeyClap/claude-forge (2 stars, last pushed 1mo ago), licensed MIT. It adds 45 tokens to every session and 752 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.
Other skills, from other repositories
ai-search-optimization
Use when building or auditing a site for SEO visibility in AI Overviews, AI Mode, or generative search (AEO / GEO / answer-engine optimization), or when someone asks how to "optimize for ChatGPT/Gemini/AI search", whether to add llms.txt or AI text files, how grounding / query fan-out / RAG pick pages, indexability…
knowledge-pipeline
Operate the local-first Weblyfe knowledge pipeline for ingesting documents and workspace content, embedding locally, semantic search, and graph exploration.
knowledge-query
Hybrid search (vector + BM25 via RRF + metadata boost) against the pgvector Knowledge base, with optional RAG synthesis. Use when the user asks factual questions that should be grounded in indexed documents (e.g., 'what do we know about X', 'search the knowledge base for Y', '@knowledge '). Pass answer=true to…
embedding-strategies
Select and optimize embedding models for semantic search and RAG applications. Use when choosing embedding models, implementing chunking strategies, or optimizing embedding quality for specific domains.
hybrid-search-implementation
Combine vector and keyword search for improved retrieval. Use when implementing RAG systems, building search engines, or when neither approach alone provides sufficient recall.
langchain-architecture
Design LLM applications using LangChain 1.x and LangGraph for agents, memory, and tool integration. Use when building LangChain applications, implementing AI agents, or creating complex LLM workflows.