The FailproofAI org's collection of agent skills - reusable, cross-agent
instruction sets that teach a coding agent how to do a specific job well.
One repo, many skills, installable with npx skills.
A skill is a folder with a SKILL.md (YAML frontmatter: name + description,
then instructions). Agents load it when a task matches the description, so the
agent gains a specialized, repeatable capability without you re-explaining it.
The format is shared across agents, so one definition works everywhere.
| Skill | What it does | Source of truth |
|---|---|---|
agenteye-cli | Operate an AgentEye deployment from the terminal via the agenteye CLI - inspect telemetry (errors, sessions, events, evals), triage alerts/incidents, manage keys/users/settings, run queries. | Synced from FailproofAI/agenteye → cli/skill/ (private). Do not hand-edit here. |
agenteye-evaluator | Put automatic quality scores on an agent's production runs - decide which dimensions are worth scoring from real sessions, scaffold the scoring service with the agenteye-evaluator Python SDK, score with rules or an LLM judge, test it against a captured session, deploy it and confirm scores land. | Synced from FailproofAI/agenteye → evaluator-sdk/skill/ (private). Do not hand-edit here. |
agenteye-python-sdk | Make an AI agent report what it did - plan which points in the agent loop to record, write the instrumentation with the agenteye Python SDK, thread session/agent identity through it, and verify the events actually land. | Synced from FailproofAI/agenteye → python-sdk/skill/ (private). Do not hand-edit here. |
failproofai-policy-author | Turn what agents keep doing wrong into enforcement for failproofai - triage a failproofai audit or an AgentEye deployment's findings, convert a CLAUDE.md/AGENTS.md into policies, or take a plain complaint ("agents keep force-pushing") and enforce it. Checks the shipped builtins and their params before writing anything, since most requests are one line of config; tests every policy it authors. | Maintained here. Not synced from anywhere - edit in this repo. |
Using the skills CLI (vercel-labs/skills). It auto-detects
your agent(s); pass -a to be explicit.
# the whole collection (project-local → ./.agents/skills/)
npx skills add FailproofAI/skills
# just one skill
npx skills add FailproofAI/skills --skill skillname
# explicitly for claude-code
npx skills add FailproofAI/skills -a claude-code
# just ONE skill - use the --skill flag (NOT a /agenteye-cli path segment)
npx skills add FailproofAI/skills --skill agenteye-cli -a claude-code
# global (all projects → ~/.claude/skills/), with real copies instead of symlinks
npx skills add FailproofAI/skills --skill agenteye-cli -a claude-code -g --copy
# Codex instead
npx skills add FailproofAI/skills --skill agenteye-cli -a codexOr point straight at the skill folder by URL:
npx skills add https://github.com/FailproofAI/skills/tree/main/skills/agenteye-cli -a claude-code
Inspect / verify / manage:
npx skills add FailproofAI/skills --list # list skills in the repo (no install)
npx skills list -a claude-code # what's installed (alias: ls)
npx skills update agenteye-cli # update an installed skill
npx skills remove agenteye-cli # remove it| Scope | Flag | Where it lands (Claude Code) |
|---|---|---|
| Project (default) | (none) | ./.claude/skills/ - commit with your project |
| Global | -g | ~/.claude/skills/ - across all projects |
By default the CLI symlinks each agent's skills dir to one canonical copy
(easy updates). If your environment doesn't follow symlinks, add --copy for
independent real copies.
Public-repo note: for anyone outside the org to
npx skills add FailproofAI/skills, this repo must be public. If it stays private, installs need auth or an internal mirror. (All three skills' content is scrubbed safe-for-public — that is a standing requirement of the sync, not a one-off.)
- Claude Code reads
.claude/skills/(project) /~/.claude/skills/(global) - not.agents/skills/..agents/skills/is the vendor-neutral path other agents use (Codex project, Cursor, Gemini CLI, …) and where the symlink canonical store lives. If the skill only shows up under~/.agents/skills/and Claude Code ignores it, the per-agent symlink wasn't created or wasn't followed. - Fix: re-install targeting the agent explicitly and forcing real copies:
npx skills add FailproofAI/skills --skill agenteye-cli -a claude-code -g --copy - Verify with the CLI, not by eyeballing dirs:
npx skills list -a claude-code. - Codex paths: project
.agents/skills/, global~/.codex/skills/.
skills/ ← this repo
├── README.md
├── LICENSE
├── CONTRIBUTING.md ← conventions + how to add a skill
├── assets/ ← banner for this README
├── templates/SKILL.template.md ← starter for a new skill (or use `npx skills init`)
├── scripts/validate-skills.py ← frontmatter/layout validator (run before merge)
└── skills/ ← one self-contained folder per skill
├── agenteye-cli/
│ ├── SKILL.md
│ ├── references/commands.md
│ └── agents/openai.yaml
├── agenteye-evaluator/
│ ├── SKILL.md
│ ├── references/
│ │ ├── scaffold.md
│ │ ├── sdk-api.md
│ │ └── session-data.md
│ └── agents/openai.yaml
├── agenteye-python-sdk/
│ ├── SKILL.md
│ ├── references/
│ │ ├── events.md
│ │ ├── install.md
│ │ └── integration.md
│ └── agents/openai.yaml
└── failproofai-policy-author/
├── SKILL.md
├── references/
│ ├── agenteye.md
│ ├── api.md
│ ├── builtins.md
│ ├── patterns.md
│ ├── rules-files.md
│ └── traps.md
├── scripts/ ← runnable helpers (test a policy, sync a reference)
└── agents/openai.yaml
Each skill folder is self-contained: its references/scripts/assets live inside
skills/<name>/ and use relative paths, because the installer copies the whole
folder and nothing outside it. See CONTRIBUTING.md to add
or update a skill.
npx skills sends anonymous usage telemetry (and transmits skill files for a
public repo); disable with DISABLE_TELEMETRY=1 or DO_NOT_TRACK=1 (auto-off in CI).
License: see LICENSE - placeholder; to be finalized before public release.