Command line

npx aiad-sdd CLI — install and tool up

The slash commands live in the agent; the aiad-sdd CLI lives in the terminal and the CI. It installs SDD Mode, keeps it up to date, and provides the deterministic verdicts (traceability, multi-runtime parity, project state) that turn rituals into measurements.

1-minute install

npx aiad-sdd init

A single command installs in your project: the .aiad/ structure (PRD, ARCHITECTURE, AGENT-GUIDE, intents, specs, metrics), the 4 Tier 1 governance agents (AI-ACT, RGPD, RGAA, RGESN) and the CRA agent shipped as a complement for software products, the 33 slash commands and the 9 skills for Claude Code and the configured CLAUDE.md file. Then, in the agent: /sdd init to start the framing.

npx aiad-sdd update — update

Updates the slash commands and the governance agents to the current version of the package, while preserving your files: your .aiad/ artifacts (PRD, Intents, SPECs, metrics) are never touched. Options: --sans-gouvernance to leave governance untouched, --dry-run for a preview, --check in CI (exit 1 if the project diverges from the package).

npx aiad-sdd status — project state

Displays the project's SDD state directly in the terminal: artifacts present, current cycle, and a maturity score from 0 to 5 — from "not initialised" to "complete". Outputs --json (machine) and --markdown (pasteable into a PR, Slack or Notion).

npx aiad-sdd trace — machine-verifiable traceability

Generates the Intent ↔ SPEC ↔ Code ↔ Tests matrix from the @intent / @spec / @verified-by / @governance annotations placed in the code. It is the final arbiter of the Drift Lock: measurement replaces an act of faith.

npx aiad-sdd trace                 # full matrix (md + json + html)
npx aiad-sdd trace --fail-on-gap   # CI: exit 1 on a blocking gap
npx aiad-sdd trace --json          # raw JSON on stdout

npx aiad-sdd emit-rules — multi-runtime

Regenerates the rule files of all the AI runtimes from the single upstream source (.aiad/AGENT-GUIDE.md + .aiad/gouvernance/): AGENTS.md, the CLAUDE.md header, .cursor/rules/*.mdc, .codex/AGENT.md, GEMINI.md and, since 1.19.0, the .kiro/steering/aiad*.md steering files. One single change in the AGENT-GUIDE, and Claude Code, Cursor, Codex, Copilot, Gemini and Kiro stay aligned.

npx aiad-sdd emit-rules                   # all targets
npx aiad-sdd emit-rules --runtime kiro    # one specific runtime
npx aiad-sdd emit-rules --check           # CI: exit 1 on divergence

Exit codes: 0 / 1 / 2

The CLI follows a three-state convention, aligned with the framework's JNSP uncertainty policy ("Je Ne Sais Pas" — I don't know):

This third code lets the CI distinguish "blocked waiting for a human" from a real bug — it is notably emitted by the cycle's deterministic scorers (research, discovery-check, mini-gate, and soon gate-precheck).

npx aiad-sdd gate-precheck — Gate pre-check (upcoming)

Next release — upcoming. Deterministic pre-check of the Execution Gate, run by /sdd gate before any SQS scoring. It reads the SPEC's "Agent execution perimeter" section outside the model; only a structured declaration can close the Gate — sensitive keywords in prose only produce a warning.

npx aiad-sdd gate-precheck <SPEC-id>          # human-readable verdict
npx aiad-sdd gate-precheck <SPEC-id> --json   # machine verdict

CI integrations — GitHub Actions

Two provided workflows lock down the SDD Mode invariants on pull requests:

Hooks

The npx aiad-sdd hooks command installs the Git pre-commit hook (and --uninstall removes it). The hooks provided in .aiad/hooks/ close the loop between the agent and the repository:

• pre-commit.sh — blocks any diff containing an unresolved TODO-JNSP: and applies the Drift Lock before commit. The git commit --no-verify bypass exists but is discouraged.
• session-start.js — executed when an agent session opens, to load the initial SDD context.
• discovery-gate.js — verifies the Discovery prerequisite (§3.5): /sdd spec and /sdd exec do not start without a linked Research at a GO / CONDITIONAL GO verdict.