Implicit Skill Routing

Rule

EXTRACTED: Agents should invoke skills by task intent, not only when the user names a skill.

For non-trivial tasks, the agent should inspect available skill metadata, read the matched SKILL.md, and follow the workflow before improvising.

Source	Use
`D:\agent-resources\SKILL-INDEX.md`	Broad shared skill catalog and routing map.
`D:\agent-resources\skills\`	Canonical shared skill library.
`.codex/skills/`	Codex project-local skills, including ARIS audit skills.
`.claude/skills/`	Claude Code / OpenCode-visible project skills.

Research audit, paper review, citation check, and experiment audit → ARIS skills.
README refresh, public repository presentation, onboarding documentation → readme-blueprint-generator.
Agent memory, tool routing, wrapper failures, or multi-agent infrastructure → agent architecture audit skills.
Skill creation or trigger/frontmatter optimization → skill-creator.
Debugging, testing, security review, frontend design, browser automation, documents, slides, and spreadsheets → search the shared skill index before acting.

Read the SKILL.md body after selecting a skill.
Use the skill’s scripts or references when they exist and are relevant.
Do not bulk-load whole skill repositories when a specific file is enough.
If no skill fits a complex task, search curated resources or public GitHub sources before inventing a new workflow.
Record useful new skill-routing knowledge in D:\agent-resources\SKILL-INDEX.md and the wiki.

AMBIGUOUS: Tiny wording edits and simple one-command tasks may not benefit from a skill; the routing cost should match task risk.
INFERRED: Better frontmatter descriptions improve passive triggering more than long always-on prompt rules.
UNVERIFIED: Some imported skills have not been fully smoke-tested on this Windows workstation and should be validated before high-stakes use.