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 wan-huiyan/agent-traffic-control --skill factcheck-subagent-needs-complete-sourcesgit clone --depth 1 https://github.com/wan-huiyan/agent-traffic-controlWrote 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/wan-huiyan/agent-traffic-control/factcheck-subagent-needs-complete-sources)<a href="https://agentmods.dev/skills/wan-huiyan/agent-traffic-control/factcheck-subagent-needs-complete-sources"><img src="https://agentmods.dev/badge/skills/wan-huiyan/agent-traffic-control/factcheck-subagent-needs-complete-sources/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/wan-huiyan/agent-traffic-control/factcheck-subagent-needs-complete-sources"><img src="https://agentmods.dev/badge/skills/wan-huiyan/agent-traffic-control/factcheck-subagent-needs-complete-sources.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 | $0.00075 | $0.01723 |
| Opus 5 | $0.00037 | $0.00861 |
| Sonnet 5 | $0.00015 | $0.00345 |
| Haiku 4.5 | $0.00007 | $0.00172 |
Grade A, and why
factcheck-subagent-needs-complete-sources 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 10d 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 — 136 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Fact-check subagents need complete primary sources
Problem
You dispatch a subagent to independently fact-check a document against its sources. To save context or tokens, you pass the agent a summary or partial dump of one source instead of the full original. The agent then reports a false positive: "Claim X is unsupported — the term appears in no source." The claim was actually fine; the supporting text was in the part of the source you abridged away.
The failure is asymmetric and dangerous: a missing source region does not make the agent say "I can't tell." It makes the agent confidently say "unsupported" or "fabricated" — because absence-of-evidence reads as evidence-of-absence when the agent believes it holds the complete source.
Context / Trigger Conditions
- About to launch an Agent/Task subagent to "fact-check", "verify", "independently check", "stress-test", or "review the claims in" a document.
- Assembling the source bundle / writing the prompt for that agent.
- A verification subagent returns "claim X unsupported / appears nowhere / fabricated / not in any source" — especially for one specific claim while everything else verifies cleanly.
Solution
Prevention — when assembling the agent's sources:
- Pass the agent the COMPLETE primary source (full file, full document dump, full repo clone), not a hand-written summary of it. If a source is large, give it the real artifact and let the agent read what it needs — do not pre-digest.
- If you genuinely must abridge, say so explicitly in the prompt: name which sources are partial and which sections were cut, and instruct the agent to mark any verdict touching those sections UNVERIFIABLE rather than INACCURATE.
- Prefer pointing the agent at retrievable originals (a cloned repo, a file path, a document ID it can fetch itself) over your transcription of them.
Response — when the agent flags a claim as unsupported:
- Before accepting an "unsupported / fabricated" verdict, read the agent's own report for hedges — a good agent will say "the dump I was given is partial" or "this section was not in my sources." That hedge is the tell.
- Reconcile the flagged claim against the COMPLETE source yourself. If the supporting text is in a region you abridged, the verdict is a false positive — keep the original claim; do not "correct" the document.
- Treat genuine errors and source-gap false positives separately: apply the real corrections, reject the false positive, and state plainly why each.
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.
- 10d ago First seen · 136 lines · 75 tokens per session scan A f1eaf5223cba
factcheck-subagent-needs-complete-sources is a skill published in the GitHub repository wan-huiyan/agent-traffic-control (3 stars, last pushed 5d ago), licensed MIT. It adds 75 tokens to every session and 1,723 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 2026-08-31.
Other skills, from other repositories
codex-coplan
Bring in an external model (Codex) to co-develop a plan — both models draft independently, align on disagreements, and merge into one plan. Load this only when the user explicitly names Codex (e.g. "let's discuss this with Codex," "what does Codex think about this approach") — don't auto-trigger just because a plan…
codex-orchestrator
Delegate execution tasks to Codex (Claude plans and reviews). Load this only when the user names Codex explicitly — no matter how big the task is, if the user hasn't mentioned Codex, do it yourself.
good-bad-ugly
Use when the user asks to "orchestrate", "conduct", invoke "good bad ugly" / "gbu", or tackle a large multi-part coding task (audits, refactor sweeps, multi-feature sessions, "clean this up and fix everything") — anything too big for one linear pass. Runs the tiered orchestration workflow: The Good (brain) plans and…
orchestrate
Use only when the user explicitly types /orchestrate:orchestrate to decompose a large task, spawn a tree of parallel worker/subplanner/verifier subagents, and collect structured handoffs. Do not invoke autonomously.
api-testing
HTTP API testing for TypeScript (Supertest) and Python (httpx, pytest). Test REST APIs, GraphQL, request/response validation, authentication, and error handling.
bun-test-coverage
Use for test coverage with Bun, --coverage flag, lcov reports, thresholds, and CI integration.