// Use when user says "let's get started", "where are we", or at beginning of a session. Reads project context from CLAUDE.md, checks git status and recent commits, and provides orientation for the session. Works across all repo types (code, research, mixed).
Convene a 3-model council (Claude + GPT via codex CLI + Gemini CLI) on a high-stakes decision. Forces cross-critique between members and surfaces where they actually disagree, breaking Claude's default agreeableness. Use when the user asks to "convene a council", "get a second opinion", "ask GPT and Gemini", "what would other models say", or has an architecture / strategy / hiring / pricing decision where being wrong is expensive. Skip for factual questions, code with one right answer, or anything premortem-shaped (route to premortem skill instead).
Architect, execute, and iterate on AI evaluations using the Data-Task-Score framework. Treats evals as the modern, quantifiable version of a PRD. Use when the user asks to "build an eval," "improve model quality," "test an agent workflow," "quantify product intuition," "move beyond vibe checks," "measure AI output," "score LLM responses," "benchmark a prompt," or "set up evaluation infrastructure." Also triggers on phrases like "how do I know if this is working," "is the model getting better," or "eval-driven development."
Apply the codebase singularity approach: reliable codebase understanding and change with a repeatable workflow, guardrails, and verification gates. Use for repo work (feature, bugfix, refactor, migration) when you need high trust, minimal diffs, and explicit validation and exit criteria.
Extract tweet series from X/Twitter profiles with full content, links, and engagement metrics. Use when: (1) User provides a tweet URL and wants similar tweets by the same author, (2) User asks to "extract tweets", "collect tweet series", or "scrape tweets", (3) User wants to analyze a user's recurring tweet format (e.g., daily updates, weekly roundups), (4) User needs tweet content with embedded links preserved for analysis. Requires Chrome browser automation tools (mcp__claude-in-chrome__*).
Use when user says "let's wrap up", "close shop", "done for today", or wants to end a session. Handles session wrap-up including git operations, documentation updates, roadmap updates, and preparing for next session. Works across all repo types.
Run a premortem on a plan, launch, product, hire, strategy, or decision — assumes it failed 6 months from now and works backward to find every reason why, then produces a revised plan. Use when the user has a concrete plan or commitment with high cost-of-being-wrong and asks to "premortem", "stress test", "find blind spots", "poke holes", "what could kill this", or "what am I missing". Skip for vague ideas without a plan, simple feedback requests, factual questions, or already-irreversible decisions.
| name | start-session |
| description | Use when user says "let's get started", "where are we", or at beginning of a session. Reads project context from CLAUDE.md, checks git status and recent commits, and provides orientation for the session. Works across all repo types (code, research, mixed). |
Orient at the beginning of a work session. Execute these steps directly (do NOT spawn a subagent — you have the session context, a subagent does not).
Check CLAUDE.md for ## Session Wrap-up, ## Deployment, sync commands, or similar sections. Note any sync/pull commands — you'll run them in Step 1.
If CLAUDE.md documents a sync or pull command (rsync, git pull, etc.), run it now. If nothing is documented, skip this step.
CLAUDE.md is already loaded into conversation context — don't re-read it. Detect repo type from type: declaration in CLAUDE.md or infer from contents (package.json = code, mostly .md = research). Only read agents.md or README.md if CLAUDE.md is missing or lacks project context.
git fetch --prune 2>/dev/null
git status --short
git branch --show-current
git log --oneline -5
After fetching, check if the current branch is behind the remote:
git rev-list --count HEAD..@{u} 2>/dev/null
If behind, note it in the summary so the user can decide whether to pull.
Also list any remote-only branches (excluding main/master) if they exist:
git branch -r --no-merged HEAD 2>/dev/null | grep -v HEAD | grep -v 'main$' | grep -v 'master$' | head -5
Check for forgotten stashes from previous sessions:
git stash list 2>/dev/null | head -5
If stashes exist, list them in the summary so the user can decide whether to pop or drop them.
If the repo uses git worktrees (common for experiment frameworks):
git worktree list 2>/dev/null
If multiple worktrees exist, list them in the summary. This surfaces sibling experiments or feature branches the user may have been working on.
Look for files in .claude/plans/. If any exist, list them with a one-line summary. These may represent in-progress work from previous sessions.
Flag any relevant background processes (pm2, nohup jobs, screen sessions, drip campaigns). These may be left over from a previous session or actively running.
Some projects use an _agent-inbox/ handoff pattern — sandboxed/container agents drop tasks there for the host Claude Code session to execute (things they can't do: deploy, touch host, run pm2, edit .env, etc.).
_agent-inbox/ in the current repo. Also check any related-repo path referenced by CLAUDE.md (e.g., a ## Related Repository section with a "Local path:" line). Auto-memory may also record an inbox path — use it if the current CLAUDE.md points there.cd <inbox-repo> && git pull..md files directly in _agent-inbox/ that are NOT in the done/ subfolder, excluding README.md (that's the inbox's own readme, not an instruction).Completion convention (for when the user chooses to process): move the file to done/ via git mv and commit the move to the inbox repo. Run any deploy command documented in the instruction.
From CLAUDE.md, extract — skip any that don't exist:
## Current Focus — Suggested starting point for the session## Roadmap — Count total vs completed items, list next 2-3 incomplete items## Session Log — Most recent entry (what was completed, what was next)If any session log entries are older than 30 days, flag them for archiving. Suggest the user move old entries to a separate file (e.g., docs/session-archive.md) or delete them to keep CLAUDE.md concise.
Provide a structured summary including: project purpose, current branch, uncommitted changes, upstream status, recent commits, current focus, roadmap progress, last session summary, existing plans, and any running background processes.
Adapt for repo type:
After the orientation, note any missing infrastructure as suggestions (don't block on them):
No CLAUDE.md — Offer to create one with project description, roadmap sections, and current focus placeholder. Ask: "What is this project for? (one sentence)"
CLAUDE.md exists but missing roadmap sections — Suggest adding ## Current Focus, ## Roadmap, and ## Session Log. These enable progress tracking and /wrap-up-session integration.
/wrap-up-session which writes the same roadmap sections this skill reads.claude/plans/, flag stale ones (>30 days) for cleanup