Help us improve
Share bugs, ideas, or general feedback.
From ai-maestro-assistant-manager-agent
Use when an approval request arrives from a peer agent and the user is unavailable. Trigger with inbound AMP approvals from COS, AUTONOMOUS, or MAINTAINER. Returns approve-autonomously / defer / escalate-to-user.
npx claudepluginhub emasoft/ai-maestro-assistant-manager-agentHow this skill is triggered — by the user, by Claude, or both
Slash command
/ai-maestro-assistant-manager-agent:amama-autonomous-fallbackamama-assistant-manager-main-agentThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Decides what to do with an inbound approval request when the user is `monitoring`, `away`, or `dnd`. Consults the 25-row reversibility matrix and the per-state authority table, applies the per-role authority overrides, and enforces the R6 v3 routing constraint (team-bound messages route via COS only). Phase 1 of TRDD-bfcedff0. The hard-floor list (always-escalate ops) is inherited from amama-am...
Creates p5.js generative art with seeded randomness, noise fields, and interactive parameter exploration. Use for algorithmic art, flow fields, or particle systems.
Share bugs, ideas, or general feedback.
Decides what to do with an inbound approval request when the user is monitoring, away, or dnd. Consults the 25-row reversibility matrix and the per-state authority table, applies the per-role authority overrides, and enforces the R6 v3 routing constraint (team-bound messages route via COS only). Phase 1 of TRDD-bfcedff0. The hard-floor list (always-escalate ops) is inherited from amama-amcos-coordination.
Apply for every approval request from a peer agent (never to user messages). Full step-by-step procedure is in references/decision-flow.md (see Resources). Summary:
away/monitoring/dnd states escalate.monitoring R-only auto; away R+C auto; dnd R-only auto; W always defer.(verdict, classification, justification, compensating_action_ref?).Phase-1 behavior: amama-presence-tracker returns unknown until phase 3 ships, so step 4 always escalates. Policy is wired but dormant.
| Verdict | When emitted | Side effect |
|---|---|---|
approve-autonomously | Matrix says R or C, state permits | Operation executed (via COS for team-internal targets); audit-log entry appended |
defer | Matrix says W, OR unclassified, OR per-role override downgraded | Reply to source with pending-ratification; audit-log entry; user reviews on return |
escalate-to-user | Hard-floor / phase-1-string-match / state ∉ {monitoring,away,dnd} / per-role W-override | Standard escalation flow |
| Error | Action |
|---|---|
| references/reversibility-matrix.md missing | Return (escalate-to-user, "matrix-absent") — strict-by-default |
| Operation key not in matrix | Return (defer, "unclassified") |
source_role not in {COS, AUTONOMOUS, MAINTAINER} | Return (defer, "unknown-source") |
decisions-pending-ratification.md unwritable | stderr warning, escalate-to-user instead of approving |
| amama-presence-tracker unreachable | Return (escalate-to-user, "presence-unreachable") |
T2 — unclassified op while away:
state = away
request = (source=COS, op="deploy-to-quantum-cloud")
verdict = (defer, "unclassified")
audit log: APPROVAL-2026-05-05-001 | Source=COS | Op=deploy-to-quantum-cloud | Verdict=defer
T10 — AUTO first-push-to-main (per-role override forces W):
state = active ; not even a fallback path
request = (source=AUTONOMOUS, op="merge-pr-squash", attrs={first_push_to_main: true})
verdict = (escalate-to-user, "AUTONOMOUS-first-push-to-main → W")
Routine R-class while away:
state = away
request = (source=COS, op="run-unit-tests", target=team-architect)
verdict = (approve-autonomously, R, "matrix-row-1")
side effect: AMP message composed to the team's COS asking COS to run-unit-tests inside the team (R6 v3 routing)
references/decision-flow.md — Full 10-step decision flow
references/reversibility-matrix.md — The 25-row classification table
TRDD-bfcedff0 (design/tasks/) — Phase-1 spec, ratification gate, change log