> ## Documentation Index
> Fetch the complete documentation index at: https://docs.whyfile.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Keep the active wrapper deferred the passive channel is suff

# Decision: Keep the active wrapper deferred; the passive channel is sufficient

**Status:** Accepted
**Date:** 2026-07-13

## Context

Does the passive AGENTS.md pointer reliably get Codex to pull and cite whyfile's recorded intent during review, or is the active focus-text wrapper (Phase 2) needed?

## Decision

End-to-end test on this repo: introduced a deliberate ADR-0016 violation (unused numpy import in src/whyfile/query.py, a real compiled constraint in intent-rules.json), installed the review-context AGENTS.md stanza, and ran a real /codex:adversarial-review via the local Codex CLI (Codex Pro, codex-cli 0.144.1). Codex read the AGENTS.md pointer, attempted 'whyfile review-context' (failed once on PATH, self-corrected to 'python -m whyfile.cli'), cross-referenced intent-rules.json, and returned verdict=needs-attention with a high-severity finding explicitly citing 'ADR-0016' and quoting the rule's own reason text, correctly recommending revert-or-supersede. This is exactly the outcome review-context was designed to produce, achieved with zero plugin coupling.

## Alternatives considered

1. Active wrapper (whyfile codex-review) guaranteeing focus-text injection
2. Passive AGENTS.md pointer only

## Recommendation

Passive AGENTS.md pointer only