☀️ Autonomous Agency Coach — Tri-Vector Delivery & Leverage Engine

Tri-vector accountability and coaching engine for high-output digital agencies, dev squads, and principals. Operates across three vectors: Team & Delivery (TDD compliance, atomic diffs, 60m blocker escalation), Client Boundaries (scope-creep interception, 48h stop-the-clock memos, anxiety-reducing commit digests), and Founder Leverage (the 70/30 leverage rule, feast-or-famine pipeline preservation, delegation matrix). Evaluates controllable inputs on a 1–10 effort rubric to eliminate performative status theater.

Modes

Support references (load on topic)

Loadable by topic; also routable as /coach-<file> commands (not separate routed modes):

• effort-rubric.md

When to Use

Trigger Conditions

Execute this skill when:

1. Daily Standup & End-of-Day Check-In (coach:team): Closing or starting a work session, scoring 5 controllable inputs (TDD, diff size, secrets, focus, blockers), and setting the tomorrow's Single Most Important Task (MIT).
2. Client Scope-Creep or Anxiety Flare (coach:client): A client asks for "just one small change", becomes anxious about progress, delays critical assets, or requires an executive progress digest.
3. Founder Burnout & Bottleneck Review (coach:founder): The agency principal is overwhelmed with tactical code fires, revenue feast-or-famine cycles occur, or deals require founder presence to close.
4. Hygiene & Verification Audit (coach:audit): Verifying that standup artifacts, evidence links, and effort logs remain pristine and secret-free.
5. Weekly Wisdom Compounding & Retrospective Synthesis: Compounding durable insights across engineering and agency domains directly into Muse Memory's ~/.memory/wiki/ and ~/.memory/USER.md (bun skills/reflection-maintenance/coach/scripts/wisdom-updater.ts --weekly-retrospective).

Anti-Triggers

Do NOT use this skill when:

• Conducting long-term quarterly strategy retreats (use periodic-retreat).
• Triaging an active P0 production incident or system breach (use incident-response).
• Executing routine pull-request code reviews (use code-review).

Quick Reference

The 5 Controllable Input Pillars (Team Mode)

┌─────────────────┬────────────────────────────────────────────────────────┐
│ Pillar          │ Controllable Execution Metric                          │
├─────────────────┼────────────────────────────────────────────────────────┤
│ 1. Testing      │ Tests written first; 100% green before commits         │
│ 2. Minimal Diff │ Zero unrelated refactoring; laser-focused changes      │
│ 3. Hygiene      │ No secrets committed; valid frontmatter and docs       │
│ 4. Focus        │ Deep work hours on Top 1 priority without distraction │
│ 5. Triage       │ Blocked tasks explicitly escalated to dead-letter      │
└─────────────────┴────────────────────────────────────────────────────────┘

The 3 Operating Vector Playbooks

Procedure

Step 1 — Select Operating Vector & Ingest Context

Match the incoming intent to team, client, founder, or audit. Load only the matched reference file from references/.

Step 2 — Execute Vector Logic

• If Team: Run coach.ts --standup. Inspect git commits from the last 24 hours (git log --since="24 hours ago"). Score each pillar 0–2 points using references/effort-rubric.md.
• If Client: Compare incoming requests against active SOW. If out of scope, draft polite change-order response. If progress requested, run coach.ts --client-digest.
• If Founder: Categorize weekly hours across Tiers 1–4. Verify that Tier 1 & 2 bandwidth $\ge 70%$. Secure 15% pipeline reserve.

Step 3 — Triage Blockers & Enforce Boundaries

• Tasks stalled for >60 minutes must be routed to dead-letter.
• Unfunded client requests must be accompanied by Phase 2 budget/timeline addenda.
• Missing client assets (>48h) trigger an automated Stop-the-Clock memo.

Step 4 — Emit Governed Artifacts

• Team: Emit daily-standup.md with 1–10 effort score and exactly 1 MIT.
• Client: Emit client progress digest or change-order memo.
• Founder: Emit weekly leverage memo and delegation assignments.

Pitfalls

• Confusing Effort with Outcome: Grading yourself poorly because an external vendor API was down, or grading high because a lucky untested hack worked. Grade the controllable inputs.
• Absorbing Scope Creep to be Polite: Giving away hours of unbilled work that erodes agency margins. Always price the change before building it.
• Founder Hero Complex: Jumping in to fix tactical CSS bugs instead of delegating to team leads and protecting strategic sales.
• Performative Chit-Chat: Turning standups into 45-minute verbal syncs without git-backed evidence.

Verification

Before finalizing any coach execution:

1. Forensic git history or SOW specifications inspected with immutable evidence.
2. Controllable inputs or leverage scores evaluated against mathematical thresholds.
3. Exactly one Single Most Important Task (MIT) or clear client boundary action declared.
4. Zero secrets or private client tokens present in emitted logs or artifacts.