docs: sync public docs to post-July harness state and capture value-proof gaps plan

[noodlemind]

Why

An external review of the harness evaluated the public docs surface (README, architecture docs, AGENTS.md) as it stood before the July hardening wave (PRs #33–#37) and concluded the system was token-heavy, unmeasured, and Copilot-bound — none of which holds at HEAD. The docs lagged reality; this PR closes that lag and captures the four gaps that survived the reconciliation.

What

Commit 1 — docs sync (4 files):

• README.md: memory-model headline updated to the episodic/semantic/behavioral tiers with a link to docs/MEMORY-MODEL.md; new "Measurement, budgets, and evals" section (CI-enforced 2 KB orient pack + ~900-token engineer budget, harness report host token metering, evals/ suite); knowledge-compounding table gains the consolidated-learnings tier; directory structure adds packages/harness/, evals/, docs/MEMORY-MODEL.md.
• docs/architecture/engineer-harness.md: new harness report telemetry paragraph; concrete budget number on the VS Code evidence row; freshness marker.
• docs/architecture/skill-driven-prompt-library.md: "Review Summary" rewritten from proposal voice to shipped-state "Standardization Summary"; @dev-kit/harness command list updated to the full shipped catalog; ## Memory Cards added to the plan-section list (aligns with _plan-template.md); freshness marker.
• AGENTS.md: budgets/report/evals/memory-model line; directory structure adds packages/, evals/, docs/MEMORY-MODEL.md; freshness marker.

Commit 2 — plan capture:

• docs/plans/2026-08-02-feat-harness-value-proof-gaps-plan.md: unlocked capture (status: open, phase: 0) of the four adopted gaps — knowledge net-benefit measurement (corpus-gated), session-level compaction, correction-effort metric, verify-tier decide-and-document — plus the reviewer's standing offer to pre-review the release-eval methodology before Terminal-Bench trials run. Each gap locks via its own planning pass before implementation.

Verification

• harness validate-plan on the new plan: pass, exit 0 (schema v1, criterion mappings, configured checks all green)
• npm --prefix packages/harness test: 655/655 pass
• No references to unpublished docs; directory listings match the real tree

Summary by CodeRabbit

• Documentation
  • Updated project guidance to describe episodic, semantic, and behavioral memory models.
  • Added documentation for context budgets, token-efficiency telemetry, deterministic evaluations, and budget-breach reporting.
  • Expanded architecture references for the harness, evaluation directories, memory cards, and skill-driven workflows.
  • Added a planning document covering knowledge measurement, context compaction, correction-effort tracking, and risk-tiered verification.
  • Updated documentation dates and standardized model-summary guidance.

[noodlemind]

@coderabbitai review

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. 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: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 63331372-72ed-4ce9-b5fa-6ab5feb63263

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

📝 Walkthrough

Walkthrough

The PR updates project and architecture documentation for the July 2026 hardening wave. It documents memory, context budgets, token telemetry, deterministic evaluations, harness commands, and a plan for remaining value-proof gaps.

Changes

Hardening documentation

Layer / File(s)	Summary
Project reference updates AGENTS.md, README.md	Project metadata, directory references, memory-model details, governance, context budgets, token reporting, and deterministic evaluations now reflect the hardening wave.
Harness and skill architecture docs/architecture/engineer-harness.md, docs/architecture/skill-driven-prompt-library.md	The architecture documents describe harness report, budget checks, token telemetry, deterministic validation, memory cards, and the expanded harness command inventory.
Value-proof gap plan docs/plans/2026-08-02-feat-harness-value-proof-gaps-plan.md	The new plan defines acceptance criteria, phased implementation work, verification requirements, risk routing, review findings, and activity history for four harness value-proof gaps.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the documentation updates and the added plan for harness value-proof gaps.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch docs-post-july-sync

---

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

[coderabbitai]

✅ Action performed

Review finished.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

[coderabbitai]

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@docs/plans/2026-08-02-feat-harness-value-proof-gaps-plan.md`:
- Line 53: Update the AC4 baseline wording in the plan to reflect the actual
enforcement mode: use “enforcement: enforce” if the baseline is fail-closed, or
explicitly identify the separate boundary meant by “fail-closed” while retaining
“enforcement: warn.” Keep the documented adopt-or-decline decision and rationale
requirement unchanged.

In `@README.md`:
- Around line 98-101: Update the opening fenced code block surrounding the
directory listing in README.md to use the text language identifier, changing the
bare fence to a text fence while preserving the listing contents and closing
fence.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: e1e3238b-bc68-4ce2-846e-844324f6400a

📥 Commits

Reviewing files that changed from the base of the PR and between 70fa8fd and 08f289d.

📒 Files selected for processing (5)

• AGENTS.md
• README.md
• docs/architecture/engineer-harness.md
• docs/architecture/skill-driven-prompt-library.md
• docs/plans/2026-08-02-feat-harness-value-proof-gaps-plan.md