Audit committed agent commands, workflows, vendored docs, and tool configuration for gaps and staleness.
npx skills add https://github.com/arndvs/ctrlshft --skill agent-assets-arndvs
Please help me install this Agent Skill. Skill: agent-assets Source: https://github.com/arndvs/ctrlshft/tree/main/skills/agent-assets Command: npx skills add https://github.com/arndvs/ctrlshft --skill agent-assets-arndvs
|| true, exit zero after failure branches, or produce no artifacts, making breakage invisible for months. - Use Case: A team notices their nightly agent review workflow has reported success for months but produced nothing. This Skill traces the silent failure, identifies stale orientation files contradicting the code, and proposes which repeated prompts should become committed slash commands. ## Quick Start Ask the agent to audit the repository's agent assets — review the .claude directory, AGENTS.md, and agent CI workflows — and report what is stale, missing, or silently failing.High-intent search queries and answers about installing and using this skill.
Inventory every command, orientation file, vendored doc, and agent workflow, then open each one and check whether it still describes the current repository. Stale assets are worse than missing ones because they produce confident wrong work without anything failing.
Three occurrences make a pattern worth committing. A committed command is identical on every run, reviewable in a PR, and fixable in one place, while retyped prose drifts invisibly. Commit the literal text that worked and generalize only when a second use case appears.
Silent failure usually comes from `|| true` on failing steps, caught exceptions that exit zero, or removed artifact uploads. The workflow reports success while producing nothing. Design so a run that did no work is distinguishable from a run that failed to do work.
Vendor docs for libraries central to the codebase, especially after breaking changes, and for anything agents fetch repeatedly. Record the source URL, version, and fetch date on each file, since unversioned copies go stale invisibly and produce confidently wrong code.
Skip autonomy when the task runs less than weekly or its output needs a human decision anyway — a committed command someone invokes is cheaper and fails visibly. Autonomy pays off only when work is repetitive, verification is automated, and a missed run costs little.
marin-communityUpdates and archives stale Markdown documentation in lib/iris, lib/zephyr, and lib/fray.
gaiastudio-aiRefreshes the GAIA Claude Code statusline runtime by re-running install-statusline.sh.
Cogni-AI-OURetrieve GitHub Actions run, job, and attempt logs to diagnose workflow failures.
sethcarltonTrack and sync personal dotfiles using a bare Git repository.