ai-startlisted

# Start ## Purpose Session welcome dashboard. The dashboard is fully rendered by a deterministic Python script (`session_bootstrap.py`). The IDE agent does no per-field derivation — it runs **exactly one command**, prints the markdown verbatim, and stops. Why this contract exists: re-probing git, sqlite, manifests, board APIs, etc. from the IDE side blows the latency budget (operator-pain #18b). The script collects every field, caches the board call (stale-while-revalidate), and emits ready-to-display markdown. Cold path: < 3 s wall (with board). Warm path: < 500 ms. ## Process Run exactly this argv — literal, no flags moved or shells added — and print its stdout verbatim: ``` uv run python .ai-engineering/scripts/session_bootstrap.py --format=markdown ``` That is the whole skill. The script is enrolled in the trusted-script lane (`hooks-manifest.json` `trustedArgvs`, D-131-12) so this exact argv bypasses RTK rewriting and IOC re-evaluation. Any other invocation form (positional flag order changes, plain `python3`, missing `--format`) falls back to the full IOC path and degrades latency. ### Hard rules - Do **not** read the manifest, run `git`, query `sqlite`, hit `gh`, glob the skills/agents tree, or count `LESSONS.md` from the agent side. The script already did all of that and embedded the result inside the markdown payload. - Do **not** rewrite the markdown the script emits. The format is the cross-IDE contract (Claude Code, Codex, Antigravity, Copilot all