Project reference
The reference for people who work on aayudha itself, or want every detail in one place. New here? Start with the README or your first hour.
Documents
- Scope and sequencing: BUILD-PLAN.md. Its Appendix B decision log up to 2026-10-17 is archived privately; deviations since are in pull request descriptions ("Deviation from the plan:"). Principles: ETHOS.md. Terms: GLOSSARY.md. Threats: threat-model.md.
- Generated: CONSTRAINTS.md (default floors, autonomy, budgets, and retention, from
internal/config/defaults.toml), LAYERS.md, and RACI.md (from the verb registry). - Install, hosts, and clouds: getting-started.md, with Azure, AWS, and Google Cloud.
- Coding agents: AGENTS.md. Contributors: CONTRIBUTING.md. Vulnerabilities:
SECURITY.md. Research notes live in the same private repository,
rahulnakmol/agentic-ayudha-research. - The guide, which the docs site is built from by
aayu tell docs --site: guide/.
Hosts
Claude Code, Codex, Cursor, and Amp are the hosts. aayu install writes each verb's skill as a real copy (never a
symlink) under .agents/skills/<verb>/SKILL.md and .claude/skills/<verb>/SKILL.md, with the MCP server and hooks.
aayu install --env sets up and checks each host's cloud agent environment, and aayu install --loops sets up the
scheduled and event loops in each host's own tools. Your agent covers each host.
Branches
| Branch | Role |
|---|---|
dev |
Default. Integration; every feature pull request targets it and squash-merges when CI is green. |
rolling |
Snapshots promoted from dev on the release schedule, tagged rolling-<date>. |
main |
Stable releases promoted from rolling; every commit is tagged v*. |
ledger |
Orphan branch holding this repository's records: Ready snapshots, plans, proof, and the actions log. |
Verbs
The tables below are generated from internal/verbs/registry.go by go generate ./.... Do not edit them by hand;
CI fails on drift. The docs site's verbs reference has
every input of every verb.
| Verb | Summary | Seat | Lane | Layers | Identity | Surfaces |
|---|---|---|---|---|---|---|
grill |
Interview a human until a work item of the given type is complete and passes validation, then write it as an issue. | Product Strategist | spec | all | bot | cli, skill, mcp |
slice |
Split a Feature into XS-L stories with observable criteria, dependencies, and a layer footprint; close XL items as split. | Product Strategist | plan | all | bot | cli, skill, mcp |
plan |
Plan a story: one checkpoint per layer, contracts, tests, instrumentation, and rollback. | Code Architect | plan | all | bot | cli, skill, mcp |
decide |
Record a Decision with options, evaluation, and recommendation; accept it, supersede it, or file a floor waiver. | Founder | spec | all | human | cli, skill, mcp |
counsel |
Consult the Founder's Sage on any topic, issue, Decision, path, or PR: the call and its door, three readings, one recommendation, cited proof, and the next aayu step; advice only, never a review. | Founder | spec | all | bot | cli, skill, mcp |
triage |
Classify intake issues as Feedback, Bug, or Incident by label only; never edits bodies or touches non-intake issues. | Product Strategist | chore | none | bot | cli, skill, mcp |
brain |
Keep the product brain: an Obsidian vault on the orphan brain branch; init, checkpoint its generated links, check, or open it. | Product Strategist | chore | none | bot | cli, skill, mcp |
build |
Build a Ready story or bug in a worktree with checkpoint commits per layer, verify each checkpoint, and open a PR. | AI Native Engineer | build | all | bot | cli, skill, mcp |
verify |
Run the deterministic floors (coverage, SAST, secrets, deps, a11y, perf) and record results in a proof pack. | AI Native Engineer | chore | all | bot | cli, skill, mcp |
inspect |
Adversarially review a PR against its plan, pack, and criteria; return a structured disposition: merge, fix, or rebuild. | Engineering Manager | review | all | bot | cli, skill, mcp |
ship |
Squash-merge a PR after a green merge gate and a human pass, write the Complete section, and clean up. | Engineering Manager | chore | all | human | cli, skill |
release |
Promote a Release along dev, rolling, main; stamp included stories released and trigger measurement. | DevOps & DevEx Tinkerer | chore | L09 | human | cli, skill, mcp |
proof |
Push, get, verify, or list proof packs on the ledger branch; comment is the merge gate's pack check. | Quality Architect | chore | all | bot | cli, skill, mcp |
ready |
Validate a work item for Ready, label it status:ready, and post the Ready snapshot; --check is read-only. | Product Strategist | chore | all | bot | cli, skill, mcp |
research |
Gather pinned design references (URLs and IDs only) and user research for a Feature. | UX Researcher & Evolutionist | spec | L01 | bot | cli, skill, mcp |
wireframe |
Produce wireframes for a Feature's states from its research and criteria. | UX Researcher & Evolutionist | build | L01 | bot | cli, skill, mcp |
mockup |
Produce design-system mockups for a Feature from its wireframes. | UX Researcher & Evolutionist | build | L01 | bot | cli, skill, mcp |
prototype |
Build a flagged, working prototype of a Feature with Code Connect mapping and a Figma screenshot diff. | UX Researcher & Evolutionist | build | L01, L02 | bot | cli, skill, mcp |
review |
Critique UI with a model lane and a deterministic lane; dispositions recapture, rebuild, fix, or ship. | UX Researcher & Evolutionist | ui-review | L01 | bot | cli, skill, mcp |
polish |
Apply a review critique to a PR's UI and re-run review. | UX Researcher & Evolutionist | build | L01 | bot | cli, skill, mcp |
baseline |
Capture visual baselines for a commit SHA. | UX Researcher & Evolutionist | chore | L01 | bot | cli, skill, mcp |
experiment |
Design an experiment for a Feature: hypothesis, metric, flag, and minimum sample size. | Revenue Architect | plan | L03, L08 | bot | cli, skill, mcp |
readout |
Read out an experiment's lift once it reaches minimum sample and write the Result. | Revenue Architect | spec | L08 | bot | cli, skill, mcp |
respond |
Respond to an incident: propose an ops catalog action, execute it only through human approval, keep the timeline. | Production Engineer | build | L07, L09 | human | cli, skill |
status |
Roll up status, health, risk, proof coverage, decisions, and cost per Goal, Initiative, and Feature. | Engineering Manager | chore | all | bot | cli, skill, mcp |
audit |
Audit floors, proof coverage, waivers, decision SLAs, autonomy exceptions, and unlogged writes; write a reproducible report. | Auditor | review | all | bot | cli, skill, mcp |
retro |
Compare estimates with observed delivery, corrections, floor misses, and cost; file findings as Feedback issues. | Engineering Manager | spec | all | bot | cli, skill, mcp |
measure |
Measure a Goal or Initiative against its baseline from analytics exports or live cloud pulls and write the Result. | Data & AI Architect | spec | L08 | bot | cli, skill, mcp |
evolve |
Turn accepted retro Feedback into a failing eval case, then a skill diff that passes it, opened as an agent-authored PR. | AI Native Engineer | build | L05 | bot | cli, skill, mcp |
brief |
Explain work to an audience in a chosen register and purpose as a grounded, accessible HTML brief. | Mentalist | spec | all | bot | cli, skill, mcp |
tell |
Write docs, guides, changelogs, release notes, launch copy, or brand copy that cite only proven Features; publish is human-only. | Marketer & Documenter | spec | L01 | human | cli, skill, mcp |
init |
Set up a product repo: labels, issue types, forms, config, ledger branch, and optionally projects, gates, identity, integrations. | Founder | chore | L09 | human | cli |
install |
Install skill copies, lock.toml, MCP registration, and hooks for each host; shows a diff unless --apply. | DevOps & DevEx Tinkerer | chore | L09 | human | cli |
doctor |
Check tools, tokens, identity, config, host installs, and existing CI; --fix re-syncs copies after showing a diff. | DevOps & DevEx Tinkerer | chore | L09 | human | cli, skill, mcp |
chart |
Chart an issue's data or AI design before planning: decisions, metrics, sources, platform, and retention; or evidence, permissions, abstention, and evals. | Data & AI Architect | spec | L05, L07, L08 | bot | cli, skill, mcp |
survey |
Survey a codebase with no product docs: index it, check an evidence-backed analysis and backlog, and capture golden outputs for parity. | Code Architect | plan | all | bot | cli, skill, mcp |
investigate |
Investigate an incident from bounded telemetry queries to the deployed revision and the implicated code, and check the note; it proposes a mitigation, never runs one. | Production Engineer | spec | all | bot | cli, skill, mcp |
slo |
Define SLOs and error budgets from the user's side, measure them from a sanitised export, and record the budget that gates release. | Production Engineer | spec | all | bot | cli, skill, mcp |
runbook |
Write and check runbooks: confirm a known failure with a bounded query, mitigate it only through the ops catalog, verify it, and roll it back. | Production Engineer | spec | all | bot | cli, skill, mcp |
scope |
Plan a release's scope in its version milestone, draft it, have a person baseline it, hand it off, assign Feedback, and hold drift to the change budget. | Product Strategist | plan | all | bot | cli, skill, mcp |
Plumbing commands
| Command | Summary | Seat |
|---|---|---|
config |
Show, validate, or initialise layered configuration. | DevOps & DevEx Tinkerer |
auth |
Show the human token and bot identity aayudha resolves, with scope warnings; set, delete, or list named secrets. | DevOps & DevEx Tinkerer |
actions |
Tail, search, or verify the hash chain of the actions log. | DevOps & DevEx Tinkerer |
models |
Show the resolved model for each lane and host. | DevOps & DevEx Tinkerer |
mcp |
Serve verbs as MCP tools over stdio. | DevOps & DevEx Tinkerer |
self-update |
Replace the binary with a signed, newer release; refuses downgrades and agents. | DevOps & DevEx Tinkerer |
upgrade |
Migrate repo config, lock, state, and skill copies to this binary's schema versions. | DevOps & DevEx Tinkerer |
version |
Print the binary version. | DevOps & DevEx Tinkerer |
halt |
Stop every agent lane, loop, and workflow in the repo until resume. | Founder |
resume |
Clear a halt. | Founder |
gc |
Prune stale worktrees and leases. | DevOps & DevEx Tinkerer |
uninit |
Remove what init created, by stored IDs. | DevOps & DevEx Tinkerer |
hook |
Entry point for host hooks; hooks call only aayu hook <name>. |
DevOps & DevEx Tinkerer |
seat |
Add, remove, or list seats and their paired humans. | Founder |
lfs-agent |
git-lfs custom transfer agent for S3/R2 storage. | DevOps & DevEx Tinkerer |
next |
Print the single best next step for this repository: the exact command to run and why. | Engineering Manager |
tour |
Guided first run: check this machine, then set the repository up at Level 1 in a few prompts; refused for agents. | Founder |
inbox |
Every human-only step waiting for you, one card at a time with the twin's counsel; Enter takes the recommendation and C clears the housekeeping. Refused for agents and needs a terminal, except --list. | Founder |
completion |
Print the shell completion script, generated from the registry, for bash, zsh, fish, or PowerShell. | DevOps & DevEx Tinkerer |