feat(inject): session-start emits synthesized prose instead of N rule lines

[Gradata]

Summary

Wires prompt_synthesizer.synthesize_brain_injection into the session-start injection hook. Fixes the autoresearch regression where the 99.2% token optimization landed on the legacy N-rules path instead of the intended synthesizer.

• Body swap in inject_brain_rules.py: the <brain-rules> block no longer contains per-rule [RULE:...] or cluster [CLUSTER:...] lines. It now carries a single slot-grouped prose block in Preston-Rhodes order (task → context → examples → persona → format → tone), with inline r:xxxx anchors preserved for capture_learning.py attribution.
• Wrapper tag unchanged: <brain-rules>...</brain-rules> still wraps the content, so middleware, handoff, sanitize, and MCP contracts do not break.
• Persona baseline: pulls from <brain_dir>/../domain/soul.md when present; otherwise empty.
• Token budget: GRADATA_SYNTH_BUDGET env var (default 400) drops lowest-priority slots first.
• Manifest unchanged: .last_injection.json still keys by 4-char anchor; the synthesizer emits the same anchors (first 4 chars of stable lesson id).

Test plan

• tests/test_cluster_injection.py — 9 tests migrated from legacy line-format assertions to the new contract (anchor count, slot labels, no [CLUSTER:]/[RULE:] prefixes). 14 pass.
• tests/test_hooks_learning.py, tests/test_lesson_applications.py, tests/test_jit_inject.py — all pass unchanged; they test the wrapper tag, not the body.
• Full suite: 3931 pass, 2 skip (pytest tests/).

Follow-up: rerun autoresearch against the synthesizer with the correct metric.

Generated with Gradata

[greptile-apps]

Your free trial has ended. If you'd like to continue receiving code reviews, you can add a payment method here.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Pro

Run ID: f7b4ddc4-90e8-41f3-817e-9911ac51fcb8

📥 Commits

Reviewing files that changed from the base of the PR and between e636bb4 and 222c524.

📒 Files selected for processing (2)

• Gradata/src/gradata/hooks/inject_brain_rules.py
• Gradata/tests/test_cluster_injection.py

---

📝 Walkthrough

Summary

• Synthesizer integration: Wired prompt_synthesizer.synthesize_brain_injection into the session-start injection hook to replace legacy per-rule line format with synthesized prose
• Body format change: <brain-rules> block now contains a single prose body in Preston-Rhodes order (task → context → examples → persona → format → tone) instead of per-rule [RULE:...] or [CLUSTER:...] lines; inline r:xxxx anchors preserved for attribution
• Persona baseline: Loaded from <brain_dir>/../domain/soul.md when available
• Token budget control: New environment variable GRADATA_SYNTH_BUDGET (default 400) controls synthesis token limits; lowest-priority slots dropped first when needed
• No breaking changes: Wrapper tag <brain-rules>...</brain-rules> unchanged; .last_injection.json schema preserved; middleware and MCP contracts unaffected
• Tests migrated: 9 tests in test_cluster_injection.py updated from legacy line-format assertions to new prose contract; all other injection tests pass unchanged
• Test results: Full suite: 3,931 pass, 2 skip

Walkthrough

Refactored brain rules injection from simple line concatenation to synthesized XML structure. New logic filters out rules covered by meta mutex, sanitizes rule descriptions for XML, injects per-rule metadata fields (rule_id, slot, example_draft, example_corrected), and applies optional persona baseline from domain/soul.md. Tests updated to validate prose-based XML output instead of legacy line formats and anchor presence using regex extraction.

Changes

Cohort / File(s)	Summary
Brain Rules Synthesis Gradata/src/gradata/hooks/inject_brain_rules.py	Replaced simple concatenation of cluster and individual lines with call to synthesize_brain_injection(). Filters out rules covered by meta mutex, sanitizes XML, constructs structured payload with per-rule fields, and loads optional persona baseline from domain/soul.md. Omits <brain-rules> entirely if synthesis input is empty.
Test Assertions Update Gradata/tests/test_cluster_injection.py	Transitioned test validations from legacy line-based formats ([RULE:], [PATTERN:], [CLUSTER:]) to prose-bounded XML structure (<brain-rules>...</brain-rules>). Replaced line-presence checks with regex anchor extraction (r:xxxx), asserts anchor survival/count in synthesized output, and strengthened metadata validation in manifest/attribution tests.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly related PRs

• feat(hooks): inject meta-rules into LLM context at session start #45: Modifies the same injection file to add <brain-meta-rules> loading and formatting, complementing the new synthesized prompt structure.
• feat: retain orchestrator + cluster-level injection #96: Directly related as it implements the original cluster-lines and individual-lines assembly logic that this PR refactors into synthesized output.

Suggested labels

feature

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/synthesizer-body-swap

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}