菜单
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".
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| 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". |
Invoke this skill after the plan has been approved (typically via Plan Mode + ExitPlanMode or /superpowers:write-plan). It serves as a checkpoint that:
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.
Scan the recent conversation for a plan. Plans come from several places:
~/.claude/plans/*.md (from Claude Code Plan Mode)/superpowers:write-planIf no plan is evident, stop and suggest:
"No plan detected. Return to Plan Mode (Opus 4.6 Max recommended) or run
/superpowers:write-planto design the implementation before starting."
If a plan exists, note its location (file path or turn reference) and proceed.
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.
Restate the 3–5 key implementation targets so the user can confirm scope before coding begins. Pull these from the plan directly:
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."
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.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
model: field changes in its markdown file, update the table here too. Keep them in sync.This skill should be used when the user asks to "security audit", "check security", "audit buddy", "check backup health", "verify backup", or "security check".
This skill should be used when the user asks to "clean cache", "cache clean", "clear build cache", "free disk space", "clean up", "remove build artifacts", "clean .build", "manage cache", or "show cache size".
This skill should be used when the user asks to "token review", "review tokens", "optimize tokens", "audit context size", "reduce token footprint", "session review", "end of session review", or "check token usage".
Pre-commit pipeline for Buddy Evolver — runs token optimization, full test-all.sh, security review, docs sync, and comment audit. INVOKE IMMEDIATELY when the user says any variant of "session review", "review session", "session end", "end session", "end the session", "ending session", "session over", "session done", "wrap up", "wrapping up", "finish up", "finishing up", "done for now", "close out", "ready to commit", or "before commit" — treat these as direct invocations, not topics for discussion. Do NOT ask whether the timing is right, do NOT argue that a commit already happened (run anyway and report what's still actionable), do NOT defer. If the user actively says "don't run" or "skip", honor that — otherwise invoke now.
This skill should be used when the user asks to "sync docs", "update docs", "sync documentation", "update readme", "update claude md", "docs out of date", "check documentation", or "fix docs".
Use after a PR is merged to GitHub main. Syncs local main, verifies the merged code builds, removes merged worktrees, and stages the current worktree for automatic self-cleanup on /exit. With --release, also bumps version, updates CHANGELOG, tags the commit, pushes to trigger marketplace publish, and verifies the local plugin reloads. Use when the user says "deploy", "session deploy", "pr merged", "post-merge sync", "deploy session", "clean up after merge", or "release new version".