Menu
Analyze brownfield codebase and create initial continuity ledger
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Run the pragmatic-skeptic 'ponytail' reviewer over the current diff to surface over-engineering — code, abstractions, dependencies, files, or process steps the problem does not need. Read-only and advisory; never trims safety, security, accessibility, data-loss handling, or their tests. Use before pushing a PR, when a change feels heavier than the task warranted, or when boilerplate/a new dependency crept in. Triggers: 'ponytail review', 'is this over-engineered', 'what can I delete here', 'did I over-build this'.
Use this skill to run a structured user-discovery interview for Caro and save the transcript into docs/discovery/transcripts/ under the project's anonymization rules. Use when a new product line or major feature spec needs to clear Gate 1 of .claude/rules/validation-discipline.md (the 20-transcripts rule). Also use to synthesize transcripts into the hypothesis ledger after each batch of 5 interviews. Triggers - "run a discovery interview for <feature>", "log this user conversation as a Caro discovery transcript", "synthesize this week's transcripts into the hypothesis ledger".
Safely reclaim disk space in the caro project. Cleans Rust build cache, stale git worktrees (both .claude/worktrees/ and .worktrees/), empty stubs, and node_modules using a tiered audit that preserves any worktree with uncommitted changes, missing remotes, or a lock file. Use when du shows the project > 30 GB or disk free is low.
DEPRECATED 2026-05-16 — use caro-shell instead. This skill recommends the --backend claude flag and a ~/.config/caro/config.toml path that do not work on the current caro 1.4.0 binary, and its 522-line educational body is 4× the size of caro-shell for no benefit to an agent. Will be removed after 2026-08-01.
Use this skill when the user needs a POSIX shell command synthesized from natural language — "how do I find/grep/awk/find files modified in the last hour", "kill the process on port 3000", "tar this up excluding .git", or any other terminal-task-as-prose. Shells out to the `caro` CLI for safety-validated command inference and presents the suggestion for explicit approval. Refuses to execute the command itself.
Build, render, ship, AND MAINTAIN the caro landing-page demo video using Remotion. Use when creating, updating, re-rendering, extending the project demo MP4 at website/public/caro-demo.mp4 — or when responding to a drift alert from the caro-demo-drift CI workflow or a beads task with label `caro-demo-video`.
| name | onboard |
| description | Analyze brownfield codebase and create initial continuity ledger |
Analyze a brownfield codebase and create an initial continuity ledger.
init-project.sh in a new projectSpawn the onboard agent:
Use the Task tool with subagent_type: "general-purpose" and this prompt:
Onboard me to this project.
Read and follow the instructions in .claude/agents/onboard.md exactly.
1. Check if thoughts/ledgers/ exists (if not, tell me to run init-project.sh)
2. Set RepoPrompt workspace to this project, then explore:
rp-cli -e "workspace switch \"$CLAUDE_PROJECT_DIR\""
rp-cli -e 'tree'
rp-cli -e 'structure .'
rp-cli -e 'builder "understand the codebase architecture"'
3. If rp-cli not available, fall back to bash (find, ls, etc.)
4. Detect tech stack
5. Ask me about my goals using AskUserQuestion
6. Create a continuity ledger at thoughts/ledgers/CONTINUITY_CLAUDE-<project>.md
The onboard process:
thoughts/ledgers/CONTINUITY_CLAUDE-<name>.md/create_plan instead/continuity_ledger