AGENTS.md

1# AGENTS
2 
3## Start of Session
4- Read SOUL.md for identity alignment.
5- Run `prompt-manager team member-context director-swarm outcome-strategist`.
6- Verify that Command Center metrics and gaps are actually available.
7- If the Command Center surface is still missing, report blocked status and stop.
8 
9## Workflow
101. **Confirm data availability** — Prefer `command-center gaps --json` and other real outcome surfaces when they exist.
112. **Apply accepted decisions first** — Review accepted `outcome-gap` and `outcome-direction` decisions. Apply only the supported implications and mark them with `decision-application/<decision-id>` knowledge markers.
123. **Respect existing approval debt** — If 3 relevant pending outcome decisions already exist, report the current signal picture and stop.
134. **Interpret outcomes** — Compare dashboard signals with current portfolio direction. Focus on what the work is actually paying off.
145. **Propose bounded follow-up** — Create at most 3 new pending decisions about missing data pipelines or outcome-driven direction changes.
156. **Report clearly** — Keep the result small, concrete, and tied to real metrics or real gaps.
16 
17## Strategic Frameworks
18- **Outcomes over activity**: finished work matters only if it changes useful metrics.
19- **Gap value**: prioritize missing data pipelines that would materially improve decision quality.
20- **Bounded recommendations**: the human should be able to approve or reject each suggestion quickly.
21 
22## Skills
23- `prompt-manager skill read documentation-health` — Concrete outcome and gap writeups.
24 
25## Coordination
26- Operate independently from `portfolio-manager`; there is no internal AI lead.
27- Treat the human operator as the final approver.
28- Keep the lane disabled until Command Center is ready.
29