Personal Codex agent skills and shared references.
Use the interactive skill manager to install or uninstall skills for one or more coding CLIs.
Requirements:
- Bash 3.2 or newer
fzf
Run it from this repository:
./bin/agent-skillsTo make agent-skills available without changing into this repository, link it into ~/.local/bin:
./bin/agent-skills --install-commandThe manager supports these CLI-specific skill directories:
| CLI | Skills directory |
|---|---|
| Codex | ${CODEX_HOME:-$HOME/.codex}/skills |
| Claude Code | $HOME/.claude/skills |
| Pi | $HOME/.pi/agent/skills |
| OpenCode | $HOME/.config/opencode/skills |
| Amp Code | $HOME/.config/amp/skills |
The manager never overwrites an occupied destination. Uninstall only removes symlinks that resolve to a skill in this repository. Because installed skills are symlinks, pulling changes into this repository updates them without a separate update command.
Run the manager tests with:
./tests/test-agent-skills.shCreates atomic git commits with clear scope.
Runs a second-model code review using Codex (gpt-5.6-sol) by default, with optional Claude (claude-fable-5) review. Pair with behavior-validator for user-visible behavior checks.
Delegates substantial implementation, fixing, code exploration, rebasing, and PR landing to Codex CLI while Claude handles design, decisions, review, and verification.
Runs Codex Security scan, diff-scan, deep-scan, finding-fix, and phase workflows for repository-wide or scoped-path security reviews.
Creates coherent Conventional Commits with concise, high-signal context. Scopes the commit from the staged index and avoids splitting one task into artificial micro-commits.
Builds distinctive, production-grade frontend interfaces with high design quality. Generates creative, polished code that avoids generic AI aesthetics.
Stress-tests plans and designs through relentless direct questioning until reaching shared understanding across every branch of the decision tree.
Surveys any codebase as a senior advisor and produces prioritized, self-contained implementation plans for other models/agents to execute. Strictly read-only on source code — never implements, fixes, or refactors anything itself.
Clones and inspects external repositories in a reusable local exploration cache (~/.explore/repos) without cluttering the active workspace.
Diagnoses runtime CPU spikes, memory growth, heap leaks, GC pressure, and OOM crashes with code-first triage, guided profiling, and before/after verification.
Finds and verifies potential sponsors for a GitHub repository by mining README sponsorship placements. Outputs a verified-leads.csv for outreach planning.
SQLite database optimization, query writing, indexing, and best practices specialist. Proactively analyzes and optimizes SQLite databases for performance and reliability.
Runs coverage suites, identifies low-coverage files, proposes high-impact tests, and confirms with the user before writing tests.
Runs an extremely strict maintainability review focused on abstraction quality, complexity, large-file risk, and spaghetti-condition growth.
Runs a manual multi-agent build workflow: scope and design on Fable, exploration on Sonnet, implementation on Opus, then review via autoreview with Fable judging and applying fixes.