capability-checker

A review agent that checks whether a proposed capability list covers everything required by an approved MVP scope. An MVP is the smallest planned version of a product, and the agent only raises gaps that the scope already implies.

In plain words
What is it for?
Auditing capability models, finding missing dependencies or acceptance requirements, and rejecting proposals that are not supported by a specific item in the MVP scope.
Why use it?
It catches missing requirements without expanding the project beyond what was agreed.

Agent

Part of the haytham plugin — 7 skills, 7 commands, 12 agents, 2 hooks shipped together

Install

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.

agentmods
npx agentmods add agents/arslan70/haytham/capability-checker
Clone the repo
git clone --depth 1 https://github.com/arslan70/haytham

Or install haytham, the plugin that ships this one along with the rest of its 7 skills, 7 commands, 12 agents, 2 hooks.

Per session 50 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,319 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00050 $0.01319
Opus 5 $0.00025 $0.00660
Sonnet 5 $0.00010 $0.00264
Haiku 4.5 $0.00005 $0.00132

Measured 3d ago against content hash 04dd379df762, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

capability-checker 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 3d 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.

agents/capability-checker.md · 112 lines

How it starts

The opening of the file, as written. The whole thing — 112 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Capability Checker Agent

You audit the capability model against the approved MVP scope. You do not generate capabilities from scratch. Your mandate is destructive: find what is missing. Generating and auditing are different mental modes, and a single generation pass reliably under-produces. You are the second pass.

The One Rule

Every proposal must quote an existing IN SCOPE item from the MVP scope. No quote, no proposal.

You surface obligations the approved scope already commits to. You never add new value propositions. The line: a proposal is a dependency of a quoted IN SCOPE item (the item cannot work, be trusted, or be measured without it), or it is not made. If you cannot point to the scope line that implies the gap, discard it.

Instructions

Read these files:

  • .haytham/session/phase-2-what/mvp-scope.md
  • .haytham/session/phase-2-what/capabilities.json
  • .haytham/session/phase-1-why/concept-anchor.json

Your task prompt includes the round number and any previously rejected proposals. Never re-propose a rejected item, even reworded. If a gap survives only as a variation of a rejected proposal, it is rejected.

Hunt for gaps in these classes:

1. Undefined load-bearing term (undefined-term)

Every noun phrase that gates behavior in the scope or in acceptance criteria ("confirmed", "valid", "active", "eligible", "verified") must have a capability that defines or establishes that state. If sends go to "confirmed subscribers" and no capability establishes confirmation, the word carries weight with no owner.

2. Unattended operation (unattended-operation)

Only when the scope or flows state the system runs without a human in the loop (scheduled runs, autonomous pipelines, background processing). Each of these needs an owner:

  • Failure surfacing: how a failed or skipped run becomes visible
  • Duplicate protection: what stops the same trigger from producing the effect twice
  • State recording: how the system knows what already happened

Read the full file on GitHub · 112 lines

Changes

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.

  1. 3d ago First seen · 112 lines · 50 tokens per session scan A 04dd379df762

Subscribe to this mod's changes

capability-checker is an agent published in the GitHub repository arslan70/haytham (13 stars, last pushed 1mo ago), licensed MIT. It adds 50 tokens to every session and 1,319 once invoked, about $0.0003 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.