akrctx-doctorlisted
Install: claude install-skill aleksandarlabs/akrcontext
# akrctx-doctor
You are the semantic Doctor workflow. The `akrctx doctor` CLI performs deterministic setup checks; your role is to interpret agent instructions, project docs, task templates, harness policy, and quality gates. Update .akrctx/wiki/ and propose instruction merges. Treat the wiki as a living artifact: add architecture patterns, conventions, testing commands, and decisions as you discover them. Do not implement product features during doctor.
## Instruction placement rubric
Classify each instruction or coherent block by load tier and discoverability. Use the narrowest tier that still loads the instruction before it is needed.
### Tiers
- **Always loaded** — `CLAUDE.md`, `AGENTS.md`, `.github/copilot-instructions.md`, and skill `description` fields. Keep only rules that are not quickly discoverable or must be known before search or action.
- **Loaded on match** — skill bodies, subagent definitions, and scoped `.github/instructions/*.instructions.md`. Put specific workflows and path rules here.
- **Loaded on invocation** — `.github/prompts/*.prompt.md` and `.claude/commands/*.md`. Check these for staleness and duplication, not length.
Treat nested `CLAUDE.md` and `AGENTS.md` as always loaded within their subtree.
### Verdicts
Assign exactly one verdict:
- **keep** — useful, not quickly discoverable, and already at the right tier.
- **move** — useful but at the wrong tier. Prefer this verdict for misplaced content.
- **delete** — already stated by code, conf