csdlc-execute

CSDLC Execute — Supervised Story Execution

Execute the stories from: $ARGUMENTS

/execute is the supervised execution mode — Steps 3, 4, and optionally 5 bundled together. Work the stories with sub-agents, review each PR, and wait for human approval before merging or moving on.

How it differs from other execution commands

Command	Scope	Human gate	Merges?
/execute	Multiple stories	Yes — waits for approval per PR	Only when told to
/shipit	Full epic	Minimal — autonomous hurricane	Yes, auto-merges between batches

/execute is for when you want to stay in the loop. /shipit is for full send.

Procedure

1. Read the stories. From the epic doc in Foundry or $ARGUMENTS:

   • Identify all stories with their dependencies
   • Build the execution order (serial vs parallel)

2. Present the plan. Show the user:

   Stories to execute:
   1. FND-E12-S1: Schema + DAOs (foundation, goes first)
   2. FND-E12-S2: GitHub OAuth (depends on S1)
   3. FND-E12-S3: .well-known endpoints (parallel with S2)
   
   I'll work each story, open a PR, and wait for your go-ahead
   before merging. Ready?
   

3. Execute each story. For each story in order: a. Craft the agent prompt (using csdlc-craft-agent-prompt skill) b. Spawn a sub-agent in an isolated worktree c. When the intern completes, run the review checklist (Step 4):

   • Scope match
   • Tests pass
   • Acceptance criteria met
   • No boundary violations d. If QA is applicable (non-trivial changes), run QA checks (Step 5) e. Write a result annotation to Foundry f. Present the PR to the user and WAIT:

   PR ready: FND-E12-S1 — Schema + DAOs
   Branch: fnd-e12-s1-schema-daos
   PR: #15
   Review: PASS (all criteria met)
   Tests: 12 new, 188 existing — all green
   
   Approve and merge? Or want to review first?
   

4. On approval: Merge the PR, move to the next story. On rejection: Note the feedback, fix or re-dispatch, present again.

5. Between stories: If the next story depends on a just-merged PR, pull main and verify the build before continuing.

6. When all stories complete: Run /present to package the full set for final review.

Parallel execution

When stories can run in parallel (no dependencies between them):

• Spawn multiple sub-agents in separate worktrees simultaneously
• Present all PRs together when the batch completes
• User can approve/reject individually

What NOT to do

• Don't auto-merge. That's /shipit territory. /execute always waits.
• Don't skip review between stories. Every PR gets the Step 4 checklist.
• Don't continue past a failed story without user input — they might want to adjust the approach.