docs(research): #148 — capturing implicit agent-authored decisions

[Knapp-Kevin]

Research-and-design deliverable for #148 (implementation explicitly out of scope per the issue). This is the decision input for Jin to pick a direction; no code ships from #148 itself.

What it covers

Evaluates the issue's 6 candidate signals + 1 added (commit-message rationale scan) against this codebase's actual substrate — scored on cost / noise / portability across interactive Claude Code, headless claude -p, and hosted runner.

Key substrate findings (grounded, not generic):

• The SessionEnd transcript that bicameral-capture-corrections (feat(skill): session-end auto-capture of uningested decisions — research + observable validation #147) already drains is the same artifact the transcript-based signals need → near-zero new substrate cost.
• The post-commit hook already exists; agent commit messages carry deliberate rationale → best signal-to-noise, fully portable (added as candidate skill: bicameral-ingest few-shot extraction (v0.4.3) #7).
• Diff-shape needs a new Edit/Write hook (not portable to headless/hosted); note_choice is a new tool; counterfactual replay is expensive / headless-only.

Recommendation (phased, portability-first)

• Phase 1 (passive): unified LLM-as-judge agent-decision lens in the existing feat(skill): session-end auto-capture of uningested decisions — research + observable validation #147 SessionEnd pass + commit-message rationale scan in the existing post-commit path. Both fully portable, no new substrate.
• Phase 2 (opt-in): bicameral.note_choice tool, backstopped by the Phase-1 scan.
• Defer: diff-shape hook (portability gap) + counterfactual replay (cost).
• Noise is the existential risk ([P0] Hard exclude dev process decisions from ledger — only track product-related decisions #393): all candidates enter as proposed behind ratification and reuse the dev-process hard-exclude.

Notes

• Research brief only (docs/research-brief-148-implicit-agent-decisions-2026-06-02.md); no code. Stands on content per the Entry Reference: desync scenario catalog #53-feat: v0.7.1 — bicameral-capture-corrections + SessionEnd hook via setup #55 research-brief precedent.
• META_LEDGER seal deferred to merge — this branch and feat(ingest): #404 Phase 1 — thread parent_decision_id through natural-format ingest #534/fix(hooks): #170 — gate the post-preflight capture reminder to non-read-only prompts #535 all fork dev tip Entry feat: v0.7.1 — bicameral-capture-corrections + SessionEnd hook via setup #55; sealing here would fork the Merkle chain.

🤖 Authored via Qor-logic SDLC on Claude Code

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: cfef8b67-aead-4a3e-896d-0c7dab1c35b1

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch research/148-implicit-agent-decisions

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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