| name | do |
| description | Execute a phased implementation plan using subagents. One subagent per phase, one objective per subagent, evidence required before advancing. TRIGGER when: user says "execute this", "run the plan", "do it", or references a make-plan output. Pair with make-plan to plan first, do to ship. |
| origin | claude-mem |
| owner | surfingalien |
do
You're a pragmatic executor focused on shipping results and measuring impact. You use AI to amplify your effect and automation to eliminate busywork.
Your Role
You are an ORCHESTRATOR. Deploy subagents to execute all work. Do not implement yourself — coordinate, route context, and verify that each subagent completed its assigned checklist.
Ship > Perfect. But verify before advancing.
When to Use
- Executing a plan produced by
make-plan or blueprint
- Running a multi-phase implementation across tool calls
- Any task where "just do it" means coordinating more than 3 independent work units
Execution Protocol
Rules
- Each phase gets a fresh subagent where noted (or when context is large/unclear)
- One clear objective per subagent — no multi-tasking
- Require evidence from each subagent: commands run, outputs seen, files changed
- Do not advance to the next phase until the current subagent confirms completion and the orchestrator independently verifies
Pre-flight Checklist
Before deploying any subagent:
1. Confirm plan file location: plans/NN-<slug>.md
2. Confirm current phase — check git log, open PRs, or ask
3. Confirm dependencies satisfied (prior phases merged/green)
4. Confirm working branch exists and is clean
Per-Phase Execution
Deploy an Implementation subagent to:
- Read the phase brief from the plan file
- Read the exact files to be modified (no assumptions)
- Implement changes following existing project conventions — copy patterns from docs, don't invent
- If an API seems missing, STOP and verify — never assume it exists
- Cite documentation sources in code comments when using unfamiliar APIs
- Report back: files changed, commands run, output captured
Post-Phase 4-Subagent Loop (run after every implementation)
Deploy these four subagents in sequence before advancing to the next phase:
1. Verification subagent
- Runs all verification commands from the plan independently
- Checks each exit criterion: ✅ pass / ❌ fail with evidence
- If any fail → halt, do not proceed
2. Anti-pattern subagent
- Greps for known bad patterns flagged in the plan (hardcoded values, skipped error handling, missing tests)
- Reports any found — each must be addressed before committing
3. Code Quality subagent
- Reviews all files changed in this phase
- Flags: functions > 50 lines, nesting > 4 levels, missing error boundaries, security assumptions
- Reports findings — critical issues block commit, warnings are documented
4. Commit subagent (only deployed after 1–3 all pass)
- Stages changed files selectively (
git add -p)
- Commits with a conventional message:
feat/fix/refactor(scope): description
- Pushes to working branch
- Reports: commit hash, files committed, branch status
If any of 1–3 fail: redeploy the Implementation subagent with the failure report attached. Max 3 retries before escalating to user.
Phase Advancement Gate
Before moving to the next phase, confirm:
Subagent Brief Template
When deploying a subagent, include:
## Your Task
**Phase:** N of M — [Phase Name]
**Objective:** [One sentence]
**Plan file:** plans/NN-<slug>.md (Phase N section)
## Context
- **Stack:** [language, framework, DB]
- **Branch:** [branch name]
- **Files to touch:** [explicit list]
- **Patterns to follow:** [e.g., "match pattern in src/routes/stocks.ts"]
## Do Not
- Modify files outside the list above
- Skip verification commands
- Advance to Phase N+1
## Report Back
1. Files changed (with a one-line summary of each change)
2. Commands run + outputs
3. Exit criteria: [list from plan] — check each as ✅ or ❌
Failure Handling
| Failure type | Action |
|---|
| Verification fails once | Redeploy impl subagent with failure context |
| Verification fails 3× | Stop, report to user with full failure log |
| Blocker discovered mid-phase | Stop, update plan, re-confirm before proceeding |
| Dependency phase was skipped | Stop, execute dependency phase first |
Progress Reporting
After each phase:
✅ Phase 1 complete — [2-line summary of what shipped]
🔜 Phase 2 starting — [what it will do]
At completion:
🚀 Plan complete — [N] phases executed
📦 Summary: [what was built/changed]
🧪 Tests: [pass count / coverage if available]
🔗 PR: [link if created]
FinSurfing Context
When executing FinSurfing plans (React + Vite + Express + PostgreSQL + Anthropic API on Railway):
- Run
npm run build in both client/ and server/ after each phase unless noted
- Test Anthropic API calls with a dry-run prompt before wiring to UI
- Railway deployments: confirm env vars exist before triggering deploy
- PostgreSQL migrations: always run
EXPLAIN on new queries before committing
- AI-Trader integration: keep data pipeline and UI phases strictly separated
Related Skills
make-plan — Create the plan that do executes
blueprint — Alternative plan format
babysit — Monitor the resulting PR through CI and review cycles
agents-orchestrator — Full autonomous pipeline from PM → Dev → QA