You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
* fix(app): reject apps referencing not-yet-runnable agents at validate/compile (#161)
The html-report agent advertised a curated `render` command + an aware-html-report
cli transport, but no such binary ships or is installable — so an app node using it
validated/compiled fine and only failed at run with "program not found".
- New agent-manifest `status: available | planned` (default available). html-report
is marked `planned` until its binary ships.
- validate_app_agents rejects (E_APP_AGENT_UNAVAILABLE) any node referencing a
`planned` agent, recursing into for-each `do:` bodies. Wired into `app validate`,
`compile_to_disk` (so no lock is produced), and the real-run pre-flight.
- `aware agent describe` shows a "planned — not yet runnable" status line so the
agent no longer looks runnable.
Note: two showcase example apps (bim-monday-audit, engineer-peer-review-delta)
reference html-report and now correctly fail validate — they were never runnable
(html-report has no binary). Tracked as a follow-up to ship html-report or revise
those examples.
* fix(app): enforce planned-agent check on dry-run + install (#161 review)
- The planned-agent pre-flight sat inside the `!dry_run` block, but a plain
`--dry-run` still dispatches read-mode nodes, so html-report.render fell through
to the spawn failure this change replaces. The check now runs for plain runs and
`--dry-run`; only `--simulate` (stubs every node, contacts no binary) skips it.
- `aware app install` ran only validate_app_safety, so an app referencing a planned
agent could install + lock despite failing validate/compile. Added
validate_app_agents to the install pre-copy validation. Found by Codex review.
0 commit comments