| name | engram-auto-coder-fairy-judgement |
| description | Decision rubric for whether to dispatch a coder-fairy at each PR-coding decision point when coder_fairy_policy is auto. Load at the delegation decision, not during implementation. |
engram-auto-coder-fairy-judgement
When to load: at each PR-coding decision point when coder_fairy_policy == "auto".
Purpose: deterministic heuristic for whether to spawn a coder-fairy or do the work directly.
Heuristic
Spawn a coder-fairy when ANY of the following conditions hold:
- Test files are touched (
tests/ directory, *_test.py, test_*.py, or equivalent). Test changes have correctness implications that benefit from a fresh-eye pass.
- More than ~100 lines of substantive code change estimated. Below this threshold, the fixed overhead (spec writing + dispatch + handoff) dominates per-line cost.
- Another fairy is already running on a stacked change. Sequencing matters; the parallel fairy may be operating on related code.
Otherwise: do the work directly. Mentally spec the change first (what touches what, what could break, what tests apply) — the spec discipline applies even when you skip the fairy spawn.
Per-invocation overrides
The user can override the heuristic in either direction:
"spawn a fairy" / "use coder-fairy" / "fairy this" → spawn (full pipeline)"do it directly" / "no fairy" / "just do it" → do directly (skip fairy)
These overrides take precedence over the heuristic for the single invocation.
Customization
Edit this file to fine-tune the heuristic for your workflow. The conditions, threshold, and override phrasings are all yours to adjust.
Lineage
Initial heuristic calibrated by empirical input. Lean: aggressive toward staying direct.
