| name | agent-turn-review |
| description | ACE's pre-send review — invokes the fleet-wide `canopy:agent-turn-review` discipline, then adds ACE's specifics. Run before EVERY outbound reply / deliverable / PR.
|
Agent turn review (ACE)
Run before every outbound action (the thing that gets dropped under load). The general discipline
is fleet-wide and DRY — invoke canopy:agent-turn-review and apply it in full:
- A. Fidelity — re-read the request, extract each ask, do EXACTLY that (read cited sources),
rate it tough.
- B. Grounded commitments — every "I'll do X" needs a concrete, executable mechanism; a vague
"sync with / coordinate with / loop in " is vapor — convert it to a runnable check or a
draft-then-ask message, or cut it.
- C. Presentation — lead with what you DID; enumerate multiple asks; link artifacts as shared
docs; verify recipients (reply-all hides
Cc: in raw views).
ACE-specifics
- Eval-skill relationship: ACE's
-qa/-eval skills grade artifacts; this is the
brief-fidelity counterpart for correspondence. (Supersedes the old skills/self-review.)
- Send path: outbound email goes ONLY via
bin/ace-email (a config/gating.json deny rail
blocks raw gog gmail send/reply as ACE; --dry-run to preview). Every send records
thread_id in the routed run's comms-log (email-communicator step 7). Turns run in review
posture: the draft is presented in-conversation and gets the human's yes before it goes.
- Multi-ask replies: when a message contained MULTIPLE asks, enumerate them in the reply and
say how each was handled, one line each (done / link / status) — the requester must see their
checklist reflected back.
- Reply mechanics follow canopy
docs/agent-operating-model.md § 1b by reference
(jjackson/ace#828): deliverables/attachments are gdocs with the draft shown inline —
never a local file the human must open; verify the recipient set from the structured thread
read (raw dumps hide Cc:) before rating fidelity complete.
- Gated in:
turn (before every reply, and the close checklist) and inbox-triage step 2d.
