| name | session-execute |
| description | Use when transitioning from Plan Mode to code execution after a plan has been approved. Prints model/effort recommendations per component and summarizes the approved plan. Use when the user says "execute session", "start building", "plan approved", "begin implementation", or "let's code". |
Session Execute — Plan → Code Transition
Invoke this skill after the plan has been approved (typically via Plan Mode + ExitPlanMode or /superpowers:write-plan). It serves as a checkpoint that:
- Confirms a plan exists to execute
- Surfaces the recommended model/effort per component
- Summarizes the approved plan for final scope confirmation
- Transitions cleanly into coding
This skill is advisory only. Claude Code has no API to switch models programmatically; the user selects the model via the Desktop App dropdown. The skill gives clear guidance so the right choice is obvious.
Step 1: Confirm a plan exists
Scan the recent conversation for a plan. Plans come from several places:
- A file under
~/.claude/plans/*.md (from Claude Code Plan Mode)
- Output of
/superpowers:write-plan
- A user-described specification earlier in the session
If no plan is evident, stop and suggest:
"No plan detected. Return to Plan Mode (Opus 4.6 Max recommended) or run /superpowers:write-plan to design the implementation before starting."
If a plan exists, note its location (file path or turn reference) and proceed.
Step 2: Display model/effort recommendations
Print this reference table verbatim:
Component Model Recommendations
═══════════════════════════════════════════════════════════
Phase Recommended Model Why
─────────────────────────────────────────────────────────
Planning Opus 4.6 Max Deep architectural reasoning
Coding Sonnet (high) Fast, accurate code generation
Testing Sonnet (high) Adequate for output parsing
─────────────────────────────────────────────────────────
Agent Model Configured in
─────────────────────────────────────────────────────────
comment-reviewer haiku agents/comment-reviewer.md
test-runner haiku agents/test-runner.md
token-reviewer haiku .claude-plugin/agents/token-reviewer.md
docs-reviewer haiku .claude-plugin/agents/docs-reviewer.md
security-reviewer sonnet agents/security-reviewer.md
cache-reviewer inherit .claude-plugin/agents/cache-reviewer.md
─────────────────────────────────────────────────────────
The agent model fields are already configured in the agent markdown files; they run at the right model automatically. The top block is the user's choice — what model the main session runs at during each phase.
Step 3: Summarize the approved plan
Restate the 3–5 key implementation targets so the user can confirm scope before coding begins. Pull these from the plan directly:
- Which files will be created
- Which files will be modified
- What verification steps will run
Keep it to a short bulleted list (≤10 bullets). If the plan is too large to summarize, show just the top-level section headers.
Then ask: "Is this the scope you want me to implement? Reply 'yes' to proceed, or describe adjustments."
Step 4: Transition message
Once scope is confirmed, print:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Phase 2: Execute
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Model: switch to Sonnet (high) in the Desktop App dropdown
if you are currently on Opus.
Next: begin implementing the plan.
Review: run /session-review when code is ready for review.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Notes for future maintenance
- This skill is purely advisory. Do not attempt to enforce model selection — Claude Code has no API for that and the Desktop App dropdown is the right interface.
- The agent model table is the source of truth. If an agent's
model: field changes in its markdown file, update the table here too. Keep them in sync.
- Keep the plan summary short. The user already read the plan once; this is just a scope-confirmation reminder, not a re-reading.
