/peec-agent

name: peec-agent
description: Orchestrator for the 5 Peec AI growth skills (peec-setup, peec-content-intel, peec-cluster, peec-outreach, peec-report). Reads project state and last learnings, detects the single highest-leverage gap, and returns exactly one next-move decision with a measurable 4-week metric — then hands off to the skill that executes it. Use when multiple priorities compete and you need one call, not a dashboard.
user-invocable: true

AI Growth Agent

Role

Pick the single next move for a Peec AI project. One decision, one skill handoff, one measurable 4-week target. Nothing else.

Input

• `project_id` — resolve via `mcp__peec-ai__list_projects` if unknown
• optional `focus` — force a specific gap type (`taxonomy|cluster|content|citation|learning`)

Output

One markdown document matching the schema in §5, plus one appended entry in `<project>/growth_loop/decisions_log.md`. Nothing else is produced.

When to use

• Weekly 30-min ritual: "what is the one thing this week?"
• When multiple priorities compete and clarity is missing
• Quarterly strategy refresh

Do not use when:

• The answer is already known — call the target skill directly
• Peec project has <7 days of data — run `peec-setup` first

---

The loop

     ┌─────────────────────────────────────────────────┐
     │                                                 │
     │  1. UNDERSTAND                                  │
     │     peec-setup  (prompts + taxonomy)   │
     │     peec-content-intel   (demand signals)       │
     │                  ↓                              │
     │  2. ANALYZE                                     │
     │     peec-cluster (zones)             │
     │     peec-content-intel      (source intel)      │
     │                  ↓                              │
     │  3. DECIDE                                      │
     │     → THIS SKILL: one zone + one action         │
     │                  ↓                              │
     │  4. EXECUTE                                     │
     │     content-write     (new content)             │
     │     peec-outreach (external citations)      │
     │                  ↓                              │
     │  5. LEARN                                       │
     │     peec-report                        │
     │                  ↓                              │
     └───────→ feedback into step 1 ────────────────────┘

One full cycle: **1 week** for active projects, **4 weeks** for maintenance.

---

Pipeline

0. Pre-flight — setup state required

Per [`_shared/SETUP_STATE.md`](../_shared/SETUP_STATE.md), this skill refuses to run without a completed setup. Before anything else:

Read <project>/growth_loop/setup_state.json
If file missing OR completed_at missing OR phases_completed lacks
   {competitors, prompts, topics, tags}:
     STOP. Output exactly:
       "No Peec setup state found at <project>/growth_loop/setup_state.json.
        Run /peec-setup first."
     Do not bootstrap setup yourself — that is peec-setup's job.
If completed_at older than 90 days: WARN once, continue.
Use peec_project_id from state — do not call list_projects again.

1. Read state — ALWAYS run /peec-checkup first

The orchestrator does not duplicate diagnostic logic. It **always** invokes `/peec-checkup` as its first substantive step and uses the resulting report as the input for gap detection in §2. This guarantees:

• Setup health (red flags) is considered before any content/citation move is recommended
• Inventory counts (brands / prompts per stage / topics / tags / chats) are explicit, not guessed
• Brand-performance snapshot is the same one the user would see if they ran the checkup themselves — no drift between agent and dashboard

# Step 1a — orchestrate the read-only diagnostic
Invoke /peec-checkup
  inputs:
    project_id    = setup_state.peec_project_id
    date_range    = last 28 days (auto-shrink if project younger)
    top_n         = 8
  output: <project>/checkups/YYYY-MM-DD_checkup.md  (skill writes this)

# Step 1b — read auxiliary history that checkup doesn't cover
Read: <project>/growth_loop/*_learnings.json      (may not exist)
Read: <project>/growth_loop/decisions_log.md      (may not exist)
Read: <project>/outreach/*_outreach_log.md        (may not exist)

# Step 1c — cross-project priors (optional, if SkillMind MCP is available)
mcp__skillmind__recall(query="<likely gap type>", k=5, filter_tags=["peec"])

If `/peec-checkup` reports `days_with_data < 7`: STOP at Tier 1 below — do not force action. If no history exists: say so plainly ("first run — deciding from checkup + structural data only, next cycle will be sharper") and continue. If SkillMind is not installed, skip the recall step — patterns are a bonus prior, not a requirement.

2. Detect gap (priority ladder — higher tier always wins)

Inputs come from the §1 checkup report — read its sections in this order:

1. **Inventory + Setup health** (checkup §0 + §1) → tiers 1 + 2 2. **Brand performance** (checkup §2) → tiers 3 + 4 + 5 3. **Top improvements ranked** (checkup §3) → confirms or overrides tier choice with concrete actions

| Tier | Gap | Signal (from checkup) | Handoff | |---|---|---|---| | 1 | Data | `days_with_data < 7` in checkup §0 | STOP. Wait, do not force action. | | 2 | Setup quality | checkup §1 Setup Health < 80% OR any P0 finding | `peec-setup` (audit or partial:<phase>) | | 3 | Taxonomy mass | <20 active prompts OR any funnel stage <2 prompts (checkup §0) | `peec-setup partial:prompts` | | 4 | Cluster | 20+ prompts, no strategic `zone:*` tags (checkup §0 tags list) | `peec-cluster` | | 5 | Content | Zones exist, but a losing hero prompt (checkup §2) has no own asset | `content-write` (external) or `peec-content-intel` for the brief | | 6 | Citation | Hero content exists, source diversity < 5 (checkup §2) OR `editorial`/`ugc` actions queued | `peec-outreach` | | 7 | Learning | Loop has been running 4+ weeks,