aayudha docs Built from de79988 (2026-10-26)
Menu

verify: run the fixed checks before the gate does

An agent saying "all tests pass" is a claim. verify runs the floors, fixed checks with fixed thresholds, on your change, and tells you what the merge gate will find, with no model deciding the result.

Field Value
In plain words Runs the fixed checks (tests, security, secrets, dependencies, accessibility, speed) and records the results.
Journey stage Prove it
Accountable seat AI Native Engineer (Nakul)
Owns the thresholds Quality Architect (Yama)
Does the work Nakul, the Accountable seat's twin, in the chore model lane, with the seat's paired human when it has one
Consulted none
Informed Engineering Manager (Drona)
Layers it can touch every layer
Who reasons no model: it is deterministic
Writes as the bot identity (a GitHub App); without one, your identity, tagged as human fallback
Where it runs the aayu command, a skill in your agent, an MCP tool (only with [mcp] expose_writes on)
The skill your agent reads skills/verify/SKILL.md

When to use it

Want a reviewer's verdict on the pull request instead? Use inspect.

Run it

Ask your agent, in the worktree that holds the change. It runs the floors once and answers for each criterion.

Use the verify skill to tell me whether issue 42 is done, criterion by criterion, and what to fix first.

aayu verify [<pr>]
Input Required What it does
<pr> no Pull request to verify; defaults to the current worktree.
--floor no Run only this floor.
--map no Only the feature-map drift check: compare every .agents/skills/verify-*/ map with the source, run nothing, and exit 3 on drift.
--check no Only the behaviour floor, on these feature-map checks: <feature>, <feature>.<check>, or all (the sweep); a run outside the merge gate is a preview.
--maintain no With --map or --check: post the maintain brief on the tracking issue for the agent app [verify] maintainer names, when the map drifted or a check failed (drift), or every time (always). One of: drift, always.
--security no Only the security floors (sast, secrets, deps, dast) on the checked-out commit, with no PR or story: what aayudha-security.yml runs on every push and nightly. With --ci, each failing finding files or updates one aayudha:security Bug.
--since no With --security: scan secrets only in the commits after this revision (a push's before); without it, or with all zeros or an unknown revision, the whole history of HEAD.

For one floor on pull request 148, from a checkout of its head: aayu verify 148 --floor coverage.

What you get back

One line per floor: its result, value, threshold, and the tool that measured it. A local proof pack, which ties each criterion to its evidence, and behaviour-check screenshots and traces in .aayudha/state/verify/. A local run is a preview: only the merge gate's pack counts.

You should see each floor as pass, then pack written to … and exit 0. Exit 3 is a floor miss: fix the cause, never the threshold. "Tool not installed"? See merge gate problems.

Needs setup: --map and --check need a feature map. Without one, verify --map says so; start it with aayu init --verify-skill and prove a behaviour change.

Gates

It checks these before it acts, and stops when one fails:

The floors themselves are the real gate. Their thresholds live in your configuration, owned by the Quality Architect, and a model cannot overrule them (floors).

What only a person does

The registry reserves no step of this verb for a person.

To ship with a floor miss, a person with the authority accepts a waiver, a recorded Decision that audit counts. Before an agent changes a feature map, it shows you the new checks and waits for a yes.

What's next