build-codebase-knowledgelisted

# Build Codebase Knowledge Maintain a compact, evidence-backed repository-intelligence layer (`context.md`, `architecture.md`, `index.json`, `manifest.json`) and run a deterministic multi-stage task resolver to locate target code with minimum token and context overhead. Source code remains authoritative. Generated knowledge is an index and navigation aid, never a substitute for verifying implementation details. ## Execution Rules 1. **Check Freshness First**: Run `python scripts/cli.py status` before initial exploration. 2. **Run Resolver for Tasks**: Execute `python scripts/cli.py resolve "<task>"` before conducting broad repository searches. 3. **Read Minimum Source Slice**: Follow the resolver's ordered read plan. Read only high-confidence candidates first. 4. **Verify Implementation**: Verify behavior and dynamic details in primary source files before making implementation decisions. 5. **Incremental Refresh After Writes**: Call `python scripts/cli.py refresh --changed-file <file>` immediately after finishing edits. 6. **No-Op Protection**: Do not regenerate fresh artifacts. Do not read every file listed in `index.json`. ## Skill Directory Resolution Execute runtime scripts with process `cwd` set to the active skill directory (containing this `SKILL.md`) or pass absolute paths: - `skill-root`: absolute path to `skills/engineering/build-codebase-knowledge/` - `repo-root`: absolute target repository path ## Workflow ### 1. Artifact Generation When artifacts are miss