Skills

AID ships 11 user-facing skills across five pipeline groups, plus one off-pipeline maintenance skill. The six numbered phases — Discover through Execute — form the mandatory sequential pipeline; every skill runs as a slash command (e.g. /aid-config) inside your AI host tool. Each entry below is generated from the skill’s own definition in canonical/skills/.

Prepare

Set up the workspace and understand the system.

aid-config

bootstrap · run once

View or update AID pipeline settings. Bare invocation shows all values in a table; first run auto-creates .aid/settings.yml from the template. Pass a dotted key (e.g., /aid-config project.name) to view + update one setting interactively.

Definition: canonical/skills/aid-config/SKILL.md

aid-discover

Phase 1 · brownfield

Brownfield project discovery with built-in quality gate. Run /aid-config first to scaffold the KB. Analyzes all repository content (code, configuration, and documentation) to populate KB documents. Reviews, collects user input, fixes issues, and gets user approval — one step per run. State-machine: GENERATE → REVIEW → Q-AND-A → FIX → APPROVAL → DONE.

Definition: canonical/skills/aid-discover/SKILL.md

aid-summarize

optional viewer

Generate a single-file knowledge-summary.html from .aid/knowledge/. Inlines Mermaid for offline diagrams, light/dark theme, click-to-expand lightbox, accessibility-first (WCAG AA). Two-grade quality gate (Machine + Human): script-verifiable checks score the Machine Grade; an interactive checklist scores the Human Grade (K1 KB-completeness, K2 fact-grounding, V1 mandatory human visual gate). APPROVAL requires BOTH grades >= minimum. Idempotent: re-running on an unchanged KB does nothing. State-machine: PREFLIGHT → STALE-CHECK → PROFILE → GENERATE → VALIDATE → MANUAL-CHECKLIST → FIX → APPROVAL → WRITEBACK → DONE.

Definition: canonical/skills/aid-summarize/SKILL.md

Define

Define the problem and how to solve it.

aid-interview

Phase 2 · TRIAGE → full or lite

Adaptive requirements gathering through conversational interview. First run builds REQUIREMENTS.md incrementally. Subsequent runs cross-reference against KB, grade, and ask targeted questions to resolve gaps and contradictions. Final step decomposes functional requirements into discrete feature files. State machine: FIRST-RUN → Q-AND-A → TRIAGE → {full: CONTINUE → COMPLETION → FEATURE-DECOMPOSITION → CROSS-REFERENCE → DONE | lite: CONDENSED-INTAKE → TASK-BREAKDOWN → LITE-REVIEW → LITE-DONE | escalated: any lite state → CONTINUE → …full path…}.

Definition: canonical/skills/aid-interview/SKILL.md

aid-specify

Phase 3 · full path only

Technical specification through conversational refinement, one feature at a time. The agent acts as a tech lead — reads KB, Requirements, and codebase, proposes technical solutions, and builds the spec collaboratively with the user. Writes to SPEC.md in the feature folder. State machine: INITIALIZE → CONTINUE → REVIEW → DONE (SPIKE / BLOCKED are loopback states that return to CONTINUE).

Definition: canonical/skills/aid-specify/SKILL.md

Map

Turn requirements into an executable task list.

aid-plan

Phase 4 · full path only

Sequence feature SPECs into deliverables — each one a functional MVP that builds on the previous. Strategy, not tactics. Use when feature SPECs are complete and you need a delivery roadmap. State machine: FIRST-RUN → REVIEW → DONE.

Definition: canonical/skills/aid-plan/SKILL.md

aid-detail

Phase 5 · full path only

Break deliverables into small, dependency-driven, typed tasks — each one a reviewable unit. The ultimate breakdown. Detects task types (RESEARCH, DESIGN, IMPLEMENT, TEST, DOCUMENT, MIGRATE, REFACTOR, CONFIGURE) from SPEC signals. One type per task. Builds execution graph per delivery with explicit dependencies and parallelism. State machine: FIRST-RUN → REVIEW → DONE.

Definition: canonical/skills/aid-detail/SKILL.md

Execute

Build, review, and test.

aid-execute

Phase 6 · 8 task types · graded loop

Execute a task based on its type: RESEARCH, DESIGN, IMPLEMENT, TEST, DOCUMENT, MIGRATE, REFACTOR, or CONFIGURE. Built-in review loop per type. State machine: EXECUTE → REVIEW → FIX → back to REVIEW → DONE when grade ≥ minimum. Branch per delivery for isolation.

Definition: canonical/skills/aid-execute/SKILL.md

Deliver (optional)

Optionally ship, monitor, and route what breaks back into the pipeline.

aid-deploy

optional · on demand

Package completed deliveries into a release. Selects eligible deliveries, verifies the combined build, packages according to project infrastructure, generates release notes, and updates artifact statuses. Use when deliveries are complete and ready to ship. State machine: IDLE → SELECTING → VERIFYING → PACKAGING → DONE.

Definition: canonical/skills/aid-deploy/SKILL.md

aid-monitor

optional · on demand

Observe production, classify findings, and route actions. Combines telemetry interpretation with triage — detect anomalies, perform root cause analysis for bugs, and route findings to aid-interview (bugs via its lite bug-fix triage; change requests as new/changed requirements). Per-work scope. Use post-deployment, on schedule, or on-demand. State machine: OBSERVE → CLASSIFY → ROUTE → DONE.

Definition: canonical/skills/aid-monitor/SKILL.md

Off-pipeline

On-demand maintenance, outside the numbered phases.

aid-housekeep

on demand

Optional on-demand housekeeping skill. Runs three gated jobs in strict order: KB-DELTA (re-discover changed docs since last KB approval) → SUMMARY-DELTA (regenerate the visual summary if the KB changed) → CLEANUP (sweep stale work-area artifacts). Each stage commits its own changes on an aid/housekeep-* branch; the skill never pushes. Re-entrant: a stalled run resumes at the stalled stage on re-invocation. State-machine: PREFLIGHT → KB-DELTA → SUMMARY-DELTA → CLEANUP → DONE.

Definition: canonical/skills/aid-housekeep/SKILL.md