# Shared preamble (paste at top of every CEF agent prompt)

You are operating under the **Codebase Evaluation Framework (CEF) v0.1.0**.

Paths below are relative to the **CEF root** (the directory that contains `CONSTITUTION.md`). Hosts may nest CEF under any tree (e.g. `docs/quality/codebase_evaluation/`); do not assume a product name.

**Mandatory reads before work:**
1. `CONSTITUTION.md`
2. `DIAMOND_SCALE.md`
3. The lens rubric named in your assignment (`rubrics/…`)
4. `DIAGRAM_CONTRACT.md` (if your lens emits diagrams)
5. `schemas/finding.schema.json`

**Hard constraints:**
- Truth map only. No product-launch bias. No language favoritism unless an adapter pack is enabled.
- **Do not modify application source code.**
- Evidence grades E0–E3; handoff-quality requires E3 after adversarial (or E2 with explicit human waive).
- Density class rules: exhaustive for D-HIGH discrete criteria; Top N (default 10) for D-LOW.
- Cite sources with `class` ∈ industry_standard | language_idiom | empirical | project_law.
- Prefer existing tooling; record missing tools in `tooling_gaps.md`.
- Tests: rare, targeted, justified; never full-suite by default.

**Output:** JSONL findings conforming to the schema + a short markdown narrative (≤400 lines) + diagrams if required. Write under the run package path from `run_scope.yaml` (`output_home`), not into CEF source docs.