- For official API references and how-to questions, use the context7 MCP tools (
resolve-library-id+query-docs) or run/c7-docs - For real-world usage patterns and debugging errors, use the gh_grep MCP tool or run
/gh-examples,/gh-fix - When both are relevant (validating docs against real usage), run
/gh-docswhich combines both
- Use Grep and Glob for text search: file structure, topic searches, config values, comments
- Use ast-grep (via
/ast,/ast-find) for precise structural queries: function definitions, call sites, imports, anti-patterns - For unfamiliar codebases, run
/repo-auth,/repo-routes,/repo-modelsto map what exists before writing code
Before running /opsx:propose on an existing codebase:
- Load the
pre-proposeskill to map auth, routes, and models in one pass - Review the summary and suggested
config.yamlseed - Save key findings to memory
- Seed
openspec/config.yamlwith architecture conventions before generating artifacts
For bug fixes, skip the full spec pipeline: use /opsx:explore to diagnose first,
then /opsx:propose fix-<name> with a minimal tasks.md (1-3 tasks).
After /opsx:apply, load the post-apply skill for a combined quality gate:
"Load the post-apply skill and review the <change-name> change."
This runs spec compliance + security scan + code quality audit and produces a go/no-go verdict.
If GO, proceed to /opsx:archive. If NO-GO, fix the flagged items first.
For a lighter review (spec compliance + code quality only, no security scan), use spec-review directly:
"Load the spec-review skill and review the <change-name> change."
For structured debugging, load the debug-investigate skill:
"Load the debug-investigate skill and debug this error: <error message>"
This chains documentation lookup, GitHub solutions, AST search, and error handling audit into a single diagnosis.
During /opsx:apply, follow red-green-refactor when tasks.md has testable acceptance criteria:
- Red — Write a failing test for the task's acceptance criteria first
- Green — Implement the minimum code to make the test pass
- Refactor — Clean up only after green, without changing behavior
This applies especially to tasks with clear input/output expectations. Skip TDD for pure config changes, migrations, or UI-only tasks where automated testing adds no value.
For features, refactoring, or debugging with more than 2 steps, use plan mode to plan before acting.