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 skills add AnthonyAlcaraz/agentic-graph-rag-skills --skill draft-tool-trust-verifiergit clone --depth 1 https://github.com/AnthonyAlcaraz/agentic-graph-rag-skillsWrote 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/anthonyalcaraz/agentic-graph-rag-skills/draft-tool-trust-verifier)<a href="https://agentmods.dev/skills/anthonyalcaraz/agentic-graph-rag-skills/draft-tool-trust-verifier"><img src="https://agentmods.dev/badge/skills/anthonyalcaraz/agentic-graph-rag-skills/draft-tool-trust-verifier/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/anthonyalcaraz/agentic-graph-rag-skills/draft-tool-trust-verifier"><img src="https://agentmods.dev/badge/skills/anthonyalcaraz/agentic-graph-rag-skills/draft-tool-trust-verifier.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00174 | $0.02342 |
| Opus 5 | $0.00087 | $0.01171 |
| Sonnet 5 | $0.00035 | $0.00468 |
| Haiku 4.5 | $0.00017 | $0.00234 |
Grade A, and why
draft-tool-trust-verifier 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 12d 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 — 186 lines — stays where its author put it; the contents beside it link to each section on GitHub.
DRAFT Tool-Trust Verifier
Overview
Tool discovery that depends on descriptions has a failure mode the chapter names directly: providers optimize descriptions for DISCOVERY, not accuracy. "Most effective solution." "Trusted by Fortune 500." "Industry-leading performance." When every tool claims to be the best, keyword-gamed descriptions defeat the retrieval algorithms — surfacing the best marketers rather than the best tools. At scale (thousands of tools) you cannot manually verify claims.
The chapter's answer is verification-based trust, on two mechanisms:
- Structured, verifiable capabilities. A tool does not get to claim it
"analyzes customer sentiment with unparalleled accuracy." It declares the
capability
sentiment_analysiswith specific input and output types that can be tested. - Performance-based trust scores. Every tool begins neutral. Successful executions raise trust; failures, high latency, or degradations lower it. The orchestrator learns to prioritize tools that are consistently reliable.
Baidu's DRAFT (Documentation Refinement through Automated Feedback and Testing) operationalizes this as a continuous learning loop that mirrors how a developer learns a new API:
- Experience Gathering — an explorer probes tool boundaries, seeks edge cases, maps failure modes, and enforces diversity to avoid redundant tests.
- Learning from Experience — analyze the gap between documentation and reality (claims "any text input" but fails on Unicode; undocumented payload-size latency). Systematic discovery of true capabilities, not error logging.
- Documentation Rewriting — generate an AI-optimized spec reflecting the discovered reality: parameter types, ranges, error conditions, real performance. Iterate until the doc converges with actual behavior.
DRAFT sidesteps the trust problem: why worry about providers gaming descriptions when your system discovers the truth anyway? This parallels Writer's gateway, which rewrites descriptions preemptively (before deployment) rather than iteratively (after observing failures) — both treat tool descriptions as an active interface, not static metadata.
What ships with it
3 files 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.
- 12d ago First seen · 186 lines · 174 tokens per session scan A c8fd906893f9
draft-tool-trust-verifier is a skill published in the GitHub repository AnthonyAlcaraz/agentic-graph-rag-skills (10 stars, last pushed 2mo ago), licensed MIT. It adds 174 tokens to every session and 2,342 once invoked, about $0.0009 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
graphify
Use for any question about a codebase, its architecture, file relationships, or project content — especially when graphify-out/ exists, where the question should be treated as a graphify query first. Turns any input (code, docs, papers, images, videos) into a persistent knowledge graph with god nodes, community…
lemmalog
Externalize working memory and logical state into the lemmalog Datalog engine (MCP). Use for ANY multi-step task where state should outlive one context window or span agents: long investigations, debugging sessions, audits, multi-agent searches, systematic explorations, planning with many interdependent constraints…
jurisd-research
Expert Australian/NZ legal research and AGLC4 citation using the jurisd MCP server. Use when finding cases or legislation (AustLII), looking up a provision offline, formatting or resolving citations, building a pinpoint, tracing who-cites-what, or producing an AGLC4 bibliography. Triggers on case law, legislation…
repo_search
Search repository text through a deterministic first-class fak command.
container-manager-kg-ingestion
Snapshot a host's Docker/Podman/Swarm inventory into the epistemic-graph knowledge graph as typed OWL nodes via the container-manager-mcp MCP server — containers, images, volumes, networks, swarm services and nodes, with their :usesImage / :runsOn / :builtFrom links. Use when the agent must record live container state…
cortex-design
Use this skill to generate well-branded interfaces and assets for Cortex, either for production or throwaway prototypes/mocks/etc. Contains essential design guidelines, colors, type, fonts, assets, and UI kit components for prototyping.