(tooling): Context lint and eval for repo harness files

## Component

Tooling / CI

## Describe the feature

Add CI-checkable **context quality gates** for repo harness files (`AGENTS.md`, `.claude/rules/`, workflow prompts): structure lint, optional LLM-as-judge eval fixtures, and drift detection—treating context with the same rigor as code.

## Use case

Repo harness files deserve a **context development lifecycle**: generate, evaluate, distribute, observe. Teams copy-paste rules into repos without tests; context rots after refactors. ABCA loads these files at runtime but does not help operators validate them.

## Proposed solution

1. `mise` task or pre-commit hook: `context:lint` (heading structure, max line length, required sections).
2. Optional `context:eval` with golden prompts + expected properties (LLM judge behind flag).
3. Docs: context retro checklist after sprints (process in operator adoption guide).
4. Future: publish blessed packs via **Agent asset registry**.

## Other information

- Complements **feature-machine-readable-engineering-standards** and **feature-context-quality-metrics**.
- Does not block task execution in v1—advisory in CI unless repo opts in.

## Acknowledgements

- [ ] I may be able to implement this feature
- [ ] This might be a breaking change

## Acceptance criteria

- [ ] Lint rules documented with examples of pass/fail.
- [ ] At least one eval fixture in repo CI for ABCA sample config.
- [ ] Guide section in `PROMPT_GUIDE.md` or new `CONTEXT_ENGINEERING.md`.

Start by reviewing AGENTS.md, .claude/rules/, and the workflow prompts, then inspect how the repo defines mise tasks or pre-commit hooks. Use the acceptance criteria to scope documented pass/fail lint rules, one ABCA sample-config eval fixture in CI, and a guide section in PROMPT_GUIDE.md or CONTEXT_ENGINEERING.md; done means these checks are advisory and opt-in.

Written by the indexing model from the issue text.