Evidence first
Scan active sessions, archives, rollout summaries, memories, Gemini/Antigravity task files, and exported transcripts before creating a new skill.
Why this exists
Skills should evolve from real work, not guesswork.
Local fit
Audit trigger fit, undertriggering, noisy activation, workflow completion, static quality, conflicts, environment assumptions, and token economics.
Public ready
Remove private paths, credentials, internal hosts, transcript quotes, one-off project facts, and platform-specific claims before release.
Three installable skills
Mine, personalize, generalize.
Mine skill-worthy workflows
Find repeated work in coding-agent history and produce evidence-backed candidates instead of brainstorming skills from scratch.
skills/skill-miner/SKILL.md
Audit and tune existing skills
Keep the original optimizer spirit: trigger fit, undertriggering, overtriggering, user reaction, workflow completion, static quality, conflicts, and P0/P1/P2 fixes.
skills/skill-personalizer/SKILL.md
Prepare skills for public release
Turn a personal skill into a portable package for GitHub, marketplaces, teams, or cross-agent sharing.
skills/skill-generalizer/SKILL.md
Install
Use the portable Agent Skills path.
Install all three skills into ~/.agents/skills, or copy them into the native directory for Codex, Claude Code, Cursor, OpenCode, Gemini CLI, or another Agent Skills-compatible agent.
git clone https://github.com/hqhq1025/skill-optimizer.git /tmp/skill-optimizer
mkdir -p ~/.agents/skills
cp -r /tmp/skill-optimizer/skills/skill-miner ~/.agents/skills/
cp -r /tmp/skill-optimizer/skills/skill-personalizer ~/.agents/skills/
cp -r /tmp/skill-optimizer/skills/skill-generalizer ~/.agents/skills/
rm -rf /tmp/skill-optimizerAgent support
Portable format, honest platform boundaries.
Codex
~/.codex/skills or .agents/skillsClaude Code
~/.claude/skills or .claude/skillsCursor
.agents/skills or .cursor/skillsOpenCode
.agents/skills, .opencode/skills, or ~/.config/opencode/skillsGemini CLIAgent Skills-compatible installs plus
GEMINI.md for always-on contextFor LLMs and crawlers
Machine-readable project facts are first-class.
The repo includes concise and full LLM context, structured metadata, a citation file, and a sitemap so AI systems can retrieve the real project facts without guessing.