skills/faildetect/SKILL.md
--- tldr: Toggleable execution mode — dry-run, verify, run, verify-or-rollback category: utility --- # /eidos:faildetect Wraps script and command execution in a verify-or-rollback loop. When active, every non-trivial bash execution follows: dry-run → verify → run → verify-or-rollback. ## Usage ``` /eidos:faildetect # enable for this session /eidos:faildetect off # disable for this session ``` ## Instructions ### 1. Route by Command Parse arguments: - No args or `o
npx skillsauth add agenticnotetaking/eidos skills/faildetectInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Wraps script and command execution in a verify-or-rollback loop. When active, every non-trivial bash execution follows: dry-run → verify → run → verify-or-rollback.
/eidos:faildetect # enable for this session
/eidos:faildetect off # disable for this session
Parse arguments:
on → enableoff → disableAnnounce the mode change:
Faildetect **enabled** — all non-trivial commands will follow the verify-or-rollback loop.
From this point forward, follow the Execution Protocol below for every bash command that modifies files, runs scripts, or produces meaningful output.
Trivial commands (e.g. ls, pwd, date, git status, git log, reading files) are exempt.
Announce the mode change:
Faildetect **disabled** — returning to normal execution.
Resume normal command execution without the protocol.
When faildetect is active, wrap each non-trivial command in these steps:
Before executing, state:
git checkout -- <files>, git reset HEAD~1, or other)If the command supports a dry-run mode or can be previewed safely:
--dry-run, --check, --whatif, or just echo the command)Run the actual command.
Compare actual output against expected outcome:
**Faildetect rollback:**
- Command: `<what was run>`
- Expected: <what should have happened>
- Actual: <what happened>
- Rollback: `<what was undone>`
- Next: <suggested fix or question for user>
Choose the lightest rollback that restores the prior state:
git checkout -- <files> (restores from last commit)git reset --soft HEAD~1 (uncommit but keep changes staged)Apply the protocol to commands that:
Skip the protocol for:
cat, ls, git log, git status, git diff)cd, pwd)tools
--- tldr: Generate a manim explainer video through staged research → outline → scenes → code → render → merge, with the outline doubling as a plan/state document category: utility --- # /eidos:video Turn a topic (or a folder of wiki-linked md) into a rendered manim video through a fixed pipeline. Pauses after each stage for review by default. Outline doubles as the plan/state document — wiki-linked checklist tracks progress. Full design in [[spec - video skill - outline driven manim pipeline
tools
--- tldr: Surface session status to the human via a native OS notification category: utility --- # /eidos:ping Send a short out-of-band signal to the human — "I'm done", "I have a question", "I hit a wall" — as a native OS notification. The agent picks the moment; the human gets a popup without watching the terminal. ## Setup (macOS, opt-in) This skill is a no-op until the human has installed the **eidos-ping** menubar app and pointed at it in `.eidos-config.yaml`: ```yaml ping_macos: /path
tools
--- tldr: Persist behaviour corrections to CLAUDE.md category: utility --- # /eidos:toclaude Update CLAUDE.md or specs to correct undesired behaviour. ## Usage ``` /eidos:toclaude [description of correction] ``` ## Instructions 1. Understand the correction — what went wrong, what should happen instead 2. Identify the right location: - **`inject/core/*.md`** — plugin rules that apply to all eidos projects (pick the matching section file) - **`inject/feature/*.md`** — rules tied to a c
development
--- tldr: Create persistent plan for multi-step work category: planning --- # /eidos:plan Create a persistent plan file for multi-step work. ## Usage ``` /eidos:plan [brief description] ``` ## Instructions ### 1. Gather Context If the target file already exists with `status: seed`, read it — its content is raw context that seeds the plan. Use the seed's notes, links, and brain dumps to draft phases and actions. Skip clarification questions the seed already covers. Search for related arti