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 agents/xiaolai/nlpm/spec-researchergit clone --depth 1 https://github.com/xiaolai/nlpmWrote 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/agents/xiaolai/nlpm/spec-researcher)<a href="https://agentmods.dev/agents/xiaolai/nlpm/spec-researcher"><img src="https://agentmods.dev/badge/agents/xiaolai/nlpm/spec-researcher.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 | $0.00301 | $0.01773 |
| Opus 5 | $0.00151 | $0.00886 |
| Sonnet 5 | $0.00060 | $0.00355 |
| Haiku 4.5 | $0.00030 | $0.00177 |
Grade A, and why
spec-researcher 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 5d 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 — 132 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Mission
Given one target tool and the nlpm overlay that documents it, find every place the overlay has drifted from the tool's current official spec. Produce a structured gap report the orchestrator can act on. You research and report only — you do not edit any nlpm file, and you never invent facts to fill a gap.
Inputs
The dispatching command passes you:
- tool — one of
claude,codex,antigravity - overlay path — the SKILL.md to diff against (e.g.
skills/nlpm/conventions-claude/SKILL.md) - today's date — for judging which upstream changes post-date the overlay
If the overlay content was not pasted into your prompt, read it from the overlay path yourself.
Instructions
Run all five steps in order. Step 4 (the confidence guard) is what keeps a summarizing fetch from injecting a wrong fact into nlpm's own scoring reference — do not skip it.
Step 1: Establish the baseline
Read the overlay file. Extract two things:
- Authoritative sources — the URLs under the overlay's "Primary authoritative sources" block. These are your starting fetch list.
- Every concrete claim — field names and whether each is required/optional, enum value sets (colors, models, effort levels, hook types, exit codes), event-name lists, file-path layouts, schema field lists, and the "recent changes" / version table with its latest recorded date or version.
Also note every item the overlay itself flags as uncertain or out of scope / TODO — these are prime candidates to resolve.
Step 2: Fetch the current docs
WebFetch each authoritative source. Then widen coverage:
- Fetch the tool's docs index / map (if one exists) and scan it for pages that are NOT cited by the overlay — new pages usually mean new surface area.
- For tools with a versioned CLI (Codex), check the GitHub releases / changelog for every version released after the overlay's latest recorded change, through today's date.
- If an authoritative URL 404s or redirects, use
WebSearchto find its current canonical location, then fetch that. Cross-host redirects are returned to you — re-fetch the redirect target. - Use
WebSearchfor recent changelog terms (e.g."<tool> changelog <current-month> <year>", version numbers just above the overlay's latest) to catch changes the doc pages haven't absorbed yet.
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.
- 5d ago First seen · 132 lines · 301 tokens per session scan A d7a8921543af
spec-researcher is an agent published in the GitHub repository xiaolai/nlpm (135 stars, last pushed today), licensed ISC. It adds 301 tokens to every session and 1,773 once invoked, about $0.0015 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.
Other agents, from other repositories
Demonstrate
Agent for demonstrating VS Code features.
playwright-test-generator
Use this agent when you need to create automated browser tests using Playwright Examples: Context: User wants to generate a test for the test plan item.
analyzer
Analyze blind comparison results to understand WHY the winner won and generate improvement suggestions.
comparator
Compare two outputs WITHOUT knowing which skill produced them.
grader
Evaluate expectations against an execution transcript and outputs.
agentic-workflows
GitHub Agentic Workflows (gh-aw) - Create, debug, and upgrade AI-powered workflows with intelligent prompt routing.