create-quiver 0.17.1 → 0.17.3

package/specs/quiver-v53-reliable-deep-project-analysis/slices/slice-01-provider-fixture-harness/CLOSURE_BRIEF.md

@@ -0,0 +1,38 @@
+# CLOSURE_BRIEF - slice-01 provider fixture harness
+
+## Status
+
+Completed on 2026-06-11.
+
+## Summary
+
+Deterministic provider-output fixtures now reproduce the observed `nika-erp` `notes` schema drift and related provider edge cases without live Codex, credentials, or a real `nika-erp` checkout.
+
+## Evidence
+
+- Added `tests/fixtures/analyze-project/provider-output-cases.json`.
+- Added command-level tests for:
+  - `nika-erp` style `notes` drift under `domain.roles`, `domain.entities`, and `domain.actions`.
+  - fenced JSON.
+  - JSON surrounded by provider text.
+  - truncated JSON.
+  - missing required fields.
+  - invalid confidence.
+  - secret-like provider output redaction.
+  - retry-success and retry-exhausted fixture sequences.
+- Invalid provider outputs assert no `.quiver` or `docs` final writes.
+
+## Validation
+
+- `node --test tests/commands/ai-analyze-project-provider.test.js` passed: 14 tests.
+- `node --test tests/commands/ai-analyze-project.test.js` passed: 3 tests.
+- `git diff --check` passed.
+
+## Decisions
+
+- Retry cases are fixture-validated in this slice, but runtime retry behavior remains deferred to `slice-05-controlled-retry-layer`.
+- Existing invalid-output behavior is preserved: invalid final JSON fails closed and writes no final docs.
+
+## Follow-ups
+
+- Use fixtures to drive `slice-03`, `slice-04`, and `slice-05`.

package/specs/quiver-v53-reliable-deep-project-analysis/slices/slice-01-provider-fixture-harness/EXECUTION_BRIEF.md

@@ -0,0 +1,60 @@
+# EXECUTION_BRIEF - slice-01 provider fixture harness
+
+## Context
+
+The live `nika-erp` run failed because Codex returned schema-invalid extra `notes` keys. CI needs deterministic fixtures that reproduce this and other likely provider drift without relying on live provider access.
+
+## Objective
+
+Create provider fake fixtures and tests for analyze-project reliability behavior.
+
+## Scope
+
+- Provider fake responses.
+- Fixture repository or fixture payloads.
+- Tests for known drift and failure modes.
+- No live provider calls.
+
+## Acceptance Criteria
+
+- `notes` drift is reproduced deterministically.
+- Tests cover fences, surrounding text, truncation, missing required fields, invalid confidence, secret-like content, retry success, and retry exhaustion.
+- Tests assert no final docs writes when final JSON is invalid.
+- Tests can run offline.
+
+## Expected Files To Modify
+
+- `tests/commands/ai-analyze-project-provider.test.js`
+- `tests/fixtures/analyze-project/**`
+- spec evidence/status files
+
+## Validations Required
+
+- `node --test tests/commands/ai-analyze-project-provider.test.js`
+- `node --test tests/commands/ai-analyze-project.test.js`
+- `git diff --check`
+
+## Risks
+
+- Fixtures that are too synthetic may not capture the real provider drift.
+- Tests that require live providers will be flaky and expensive.
+
+## Dependencies
+
+- Depends on `slice-00-analysis-run-contract`.
+
+## Instructions For Executor
+
+1. Build the smallest useful fixture set.
+2. Keep provider fake output realistic enough to catch the observed drift.
+3. Avoid introducing network or credential requirements.
+
+## Completion Checklist
+
+- [ ] Fixtures added.
+- [ ] Tests added or updated.
+- [ ] Closure brief records failing/passing behavior.
+
+## Conditions Of Closure
+
+- The reliability bug can be reproduced without live Codex.

package/specs/quiver-v53-reliable-deep-project-analysis/slices/slice-01-provider-fixture-harness/slice.json

@@ -0,0 +1,74 @@
+{
+  "slice_id": "slice-01-provider-fixture-harness",
+  "ticket": "QUIVER-53-01",
+  "type": "test-infrastructure",
+  "title": "Provider fixture harness",
+  "objective": "Create deterministic provider-output fixtures that reproduce analyze-project drift without live provider dependency.",
+  "description": "Adds tests and fixtures for provider JSON drift cases including extra keys, markdown fences, truncation, missing fields, invalid confidence, secret-like content, retry success, and retry exhaustion.",
+  "git": {
+    "branch_type": "feature",
+    "base_branch": "main",
+    "branch_slug": "v53-provider-fixture-harness",
+    "branch_name": "feature/QUIVER-53-01-v53-provider-fixture-harness"
+  },
+  "files": [
+    "tests/commands/ai-analyze-project-provider.test.js",
+    "tests/fixtures/analyze-project/**",
+    "specs/quiver-v53-reliable-deep-project-analysis/STATUS.md",
+    "specs/quiver-v53-reliable-deep-project-analysis/EVIDENCE_REPORT.md",
+    "specs/quiver-v53-reliable-deep-project-analysis/slices/slice-01-provider-fixture-harness/CLOSURE_BRIEF.md",
+    "specs/quiver-v53-reliable-deep-project-analysis/slices/slice-01-provider-fixture-harness/slice.json"
+  ],
+  "expected_read_paths": [
+    "tests/commands/ai-analyze-project-provider.test.js",
+    "tests/commands/ai-analyze-project.test.js",
+    "src/create-quiver/commands/ai.js",
+    "src/create-quiver/lib/ai/providers.js",
+    "src/create-quiver/lib/ai/analyze-project-parser.js",
+    "src/create-quiver/lib/ai/analyze-project-schema.js"
+  ],
+  "allowed_write_paths": [
+    "tests/commands/ai-analyze-project-provider.test.js",
+    "tests/fixtures/analyze-project/**",
+    "specs/quiver-v53-reliable-deep-project-analysis/STATUS.md",
+    "specs/quiver-v53-reliable-deep-project-analysis/EVIDENCE_REPORT.md",
+    "specs/quiver-v53-reliable-deep-project-analysis/slices/slice-01-provider-fixture-harness/CLOSURE_BRIEF.md",
+    "specs/quiver-v53-reliable-deep-project-analysis/slices/slice-01-provider-fixture-harness/slice.json"
+  ],
+  "depends_on": [
+    "slice-00-analysis-run-contract"
+  ],
+  "parallel_safe": "never",
+  "parallel_safe_reason": "The fixture harness defines deterministic reproduction used by repair and retry slices.",
+  "must": [
+    "Add provider fake fixtures for the observed nika-erp notes drift.",
+    "Add fixtures for fences, surrounding text, truncation, missing required fields, invalid confidence, and secret-like output.",
+    "Make tests deterministic and independent from live Codex.",
+    "Ensure fixtures assert no final docs writes on invalid output."
+  ],
+  "not_included": [
+    "Live provider benchmarking as a required CI gate.",
+    "Runtime repair implementation beyond test scaffolding.",
+    "Structural map implementation."
+  ],
+  "acceptance": [
+    "A fixture reproduces unsupported notes keys under domain roles/entities/actions.",
+    "Provider fake tests fail before reliability implementation and pass after later slices.",
+    "Tests can run without network or provider credentials.",
+    "No live nika-erp checkout is required for CI."
+  ],
+  "tests": [
+    "node --test tests/commands/ai-analyze-project-provider.test.js",
+    "node --test tests/commands/ai-analyze-project.test.js",
+    "git diff --check"
+  ],
+  "validation_hints": [
+    "Keep live provider execution out of deterministic tests."
+  ],
+  "estimated_hours": 4,
+  "actual_hours": 2,
+  "status": "completed",
+  "started_at": "2026-06-11T12:00:00Z",
+  "completed_at": "2026-06-11T12:03:13Z",
+  "blocked_reason": null
+}

package/specs/quiver-v53-reliable-deep-project-analysis/slices/slice-02-safe-context-boundary/CLOSURE_BRIEF.md

@@ -0,0 +1,33 @@
+# CLOSURE_BRIEF - slice-02 safe context boundary
+
+## Status
+
+Completed on 2026-06-11.
+
+## Summary
+
+Provider-bound analyze-project context is now framed as untrusted data, redacted before provider execution, and persisted as a selected-context manifest for audit runs. `--dry-run` remains no-provider and no-write.
+
+## Evidence
+
+- `analyze-project` prompts now include explicit untrusted repository data delimiters and instructions not to follow repository-content instructions.
+- Provider executions write `.quiver/runs/run-.../context/selected-context.json`.
+- Context manifests include selected files, prompt bytes, truncation, read errors, safety exclusions, privacy preflight status, and safety boundary flags.
+- Provider tests assert secret-like content is redacted before prompt/provider handoff and manifests do not contain secrets.
+
+## Validation
+
+- `node --test tests/commands/ai-analyze-project-provider.test.js` passed: 14 tests.
+- `node --test tests/commands/ai-analyze-project.test.js` passed: 3 tests.
+- `node --test tests/lib/ai-safety.test.js` passed: 4 tests.
+- `npm run smoke:create-quiver` passed.
+- `git diff --check` passed.
+
+## Decisions
+
+- Real provider executions now create audit artifacts under `.quiver/runs`; this does not count as final docs/product-code writes.
+- `--print-prompt` remains no-run and does not create the selected-context manifest.
+
+## Follow-ups
+
+- Use the safe context manifest in audit and review transaction slices.

package/specs/quiver-v53-reliable-deep-project-analysis/slices/slice-02-safe-context-boundary/EXECUTION_BRIEF.md

@@ -0,0 +1,68 @@
+# EXECUTION_BRIEF - slice-02 safe context boundary
+
+## Context
+
+Provider output is not the only untrusted input. Repository content can contain secrets or prompt-injection instructions. Quiver must redact and frame context before sending it to a provider.
+
+## Objective
+
+Add a safe context boundary before provider execution.
+
+## Scope
+
+- Pre-provider redaction.
+- Prompt-injection guardrails.
+- Selected-context manifest.
+- Existing path exclusions and size budgets.
+- `--dry-run` no-write preservation.
+
+## Acceptance Criteria
+
+- Secret-like content is redacted before provider execution.
+- Repository content is delimited and described as data, not instructions.
+- Unsafe paths are still excluded.
+- Provider runs save a selected-context manifest.
+- `--dry-run` writes 0 files.
+
+## Expected Files To Modify
+
+- `src/create-quiver/lib/ai/analyze-project-discovery.js`
+- `src/create-quiver/lib/ai/analyze-project-sampling.js`
+- `src/create-quiver/lib/ai/analyze-project-prompts.js`
+- `src/create-quiver/lib/ai/safety.js`
+- analyze-project tests/fixtures
+
+## Validations Required
+
+- `node --test tests/commands/ai-analyze-project.test.js`
+- `node --test tests/commands/ai-analyze-project-provider.test.js`
+- `npm run smoke:create-quiver`
+- `git diff --check`
+
+## Risks
+
+- Redaction after provider execution is too late.
+- Over-redaction can reduce analysis quality.
+- Prompt-injection guardrails can be weakened if repo content is interpolated carelessly.
+
+## Dependencies
+
+- Depends on `slice-00-analysis-run-contract`.
+- Depends on `slice-01-provider-fixture-harness`.
+
+## Instructions For Executor
+
+1. Keep redaction deterministic and testable.
+2. Do not rely on provider obedience for safety.
+3. Keep the context manifest free of unredacted secret values.
+
+## Completion Checklist
+
+- [ ] Redaction tests pass.
+- [ ] Prompt boundary is explicit.
+- [ ] Dry-run no-write behavior is preserved.
+- [ ] Closure brief records evidence.
+
+## Conditions Of Closure
+
+- Quiver can prove that selected context is safe before provider execution.

package/specs/quiver-v53-reliable-deep-project-analysis/slices/slice-02-safe-context-boundary/slice.json

@@ -0,0 +1,85 @@
+{
+  "slice_id": "slice-02-safe-context-boundary",
+  "ticket": "QUIVER-53-02",
+  "type": "security",
+  "title": "Safe context boundary",
+  "objective": "Ensure analyze-project redacts and bounds context before provider execution.",
+  "description": "Adds the pre-provider safety boundary for selected context, including prompt-injection guardrails, secret redaction before sending, size limits, selected-context manifests, and safety exclusions.",
+  "git": {
+    "branch_type": "feature",
+    "base_branch": "main",
+    "branch_slug": "v53-safe-context-boundary",
+    "branch_name": "feature/QUIVER-53-02-v53-safe-context-boundary"
+  },
+  "files": [
+    "src/create-quiver/lib/ai/analyze-project-discovery.js",
+    "src/create-quiver/lib/ai/analyze-project-sampling.js",
+    "src/create-quiver/lib/ai/analyze-project-prompts.js",
+    "src/create-quiver/lib/ai/safety.js",
+    "tests/commands/ai-analyze-project.test.js",
+    "tests/commands/ai-analyze-project-provider.test.js",
+    "specs/quiver-v53-reliable-deep-project-analysis/STATUS.md",
+    "specs/quiver-v53-reliable-deep-project-analysis/EVIDENCE_REPORT.md",
+    "specs/quiver-v53-reliable-deep-project-analysis/slices/slice-02-safe-context-boundary/CLOSURE_BRIEF.md",
+    "specs/quiver-v53-reliable-deep-project-analysis/slices/slice-02-safe-context-boundary/slice.json"
+  ],
+  "expected_read_paths": [
+    "src/create-quiver/lib/ai/analyze-project-discovery.js",
+    "src/create-quiver/lib/ai/analyze-project-sampling.js",
+    "src/create-quiver/lib/ai/analyze-project-prompts.js",
+    "src/create-quiver/lib/ai/safety.js",
+    "src/create-quiver/commands/ai.js"
+  ],
+  "allowed_write_paths": [
+    "src/create-quiver/lib/ai/analyze-project-discovery.js",
+    "src/create-quiver/lib/ai/analyze-project-sampling.js",
+    "src/create-quiver/lib/ai/analyze-project-prompts.js",
+    "src/create-quiver/lib/ai/safety.js",
+    "tests/commands/ai-analyze-project.test.js",
+    "tests/commands/ai-analyze-project-provider.test.js",
+    "tests/fixtures/analyze-project/**",
+    "specs/quiver-v53-reliable-deep-project-analysis/STATUS.md",
+    "specs/quiver-v53-reliable-deep-project-analysis/EVIDENCE_REPORT.md",
+    "specs/quiver-v53-reliable-deep-project-analysis/slices/slice-02-safe-context-boundary/CLOSURE_BRIEF.md",
+    "specs/quiver-v53-reliable-deep-project-analysis/slices/slice-02-safe-context-boundary/slice.json"
+  ],
+  "depends_on": [
+    "slice-00-analysis-run-contract",
+    "slice-01-provider-fixture-harness"
+  ],
+  "parallel_safe": "conditional",
+  "parallel_safe_reason": "Can run beside schema grouping only if files do not overlap.",
+  "must": [
+    "Redact secret-like content before provider execution.",
+    "Treat repository content as untrusted data in prompts.",
+    "Keep existing unsafe path exclusions.",
+    "Write a selected-context manifest during provider runs.",
+    "Ensure --dry-run still writes 0 files."
+  ],
+  "not_included": [
+    "Schema repair.",
+    "Retry behavior.",
+    "Final docs writes."
+  ],
+  "acceptance": [
+    "Secret-like fixture content is not sent to the provider fake unredacted.",
+    "Prompt text clearly separates instructions from repository data.",
+    "Selected context manifest records files, bytes, truncation, and exclusions.",
+    "--dry-run remains no-write and no-provider."
+  ],
+  "tests": [
+    "node --test tests/commands/ai-analyze-project.test.js",
+    "node --test tests/commands/ai-analyze-project-provider.test.js",
+    "npm run smoke:create-quiver",
+    "git diff --check"
+  ],
+  "validation_hints": [
+    "Test redaction before provider execution, not only before artifact persistence."
+  ],
+  "estimated_hours": 6,
+  "actual_hours": 4,
+  "status": "completed",
+  "started_at": "2026-06-11T12:07:10Z",
+  "completed_at": "2026-06-11T12:11:24Z",
+  "blocked_reason": null
+}

package/specs/quiver-v53-reliable-deep-project-analysis/slices/slice-03-schema-error-grouping/CLOSURE_BRIEF.md

@@ -0,0 +1,34 @@
+# CLOSURE_BRIEF - slice-03 schema error grouping
+
+## Status
+
+Completed on 2026-06-11.
+
+## Summary
+
+Provider analysis schema failures now render as grouped, bounded, actionable CLI output while complete validation details are preserved in `.quiver/runs/.../validation/analyze-project-validation.json`.
+
+## Evidence
+
+- Added grouped issue summaries by issue type and path family.
+- Added cause hints and validation manifest path to provider validation errors.
+- Added validation manifest persistence for invalid provider analysis JSON.
+- Updated provider tests to assert compact `notes` drift output and complete manifest contents.
+
+## Validation
+
+- `node --test tests/commands/ai-analyze-project-provider.test.js` passed: 14 tests.
+- `node --test tests/commands/ai-analyze-project.test.js` passed: 3 tests.
+- `node --test tests/lib/ai-analyze-project-parser.test.js` passed: 5 tests.
+- `node --test tests/lib/ai-analyze-project-validation.test.js` passed: 4 tests.
+- `npm run docs:check` passed.
+- `git diff --check` passed.
+
+## Decisions
+
+- Invalid provider JSON may now create an audit validation manifest under `.quiver/runs`; final docs and product code still remain unchanged.
+- Schema validation remains strict. This slice does not repair or retry invalid JSON.
+
+## Follow-ups
+
+- `slice-04-safe-repair-layer` will use grouped validation details to decide repairability.

package/specs/quiver-v53-reliable-deep-project-analysis/slices/slice-03-schema-error-grouping/EXECUTION_BRIEF.md

@@ -0,0 +1,65 @@
+# EXECUTION_BRIEF - slice-03 schema error grouping
+
+## Context
+
+The current provider failure can print many repeated schema issues. Users need a compact explanation and a complete manifest for debugging.
+
+## Objective
+
+Group schema errors into actionable console output while preserving complete validation evidence.
+
+## Scope
+
+- Error grouping.
+- Cause hints.
+- Next safe command.
+- Validation manifest.
+- i18n messages.
+
+## Acceptance Criteria
+
+- Repeated `notes` errors are grouped.
+- Console output shows a bounded number of examples.
+- Complete validation issues are available in a manifest.
+- `--json` remains machine-readable.
+
+## Expected Files To Modify
+
+- `src/create-quiver/commands/ai.js`
+- `src/create-quiver/lib/ai/analyze-project-validation.js`
+- `src/create-quiver/lib/i18n/messages/en.js`
+- `src/create-quiver/lib/i18n/messages/es.js`
+- provider tests/fixtures
+
+## Validations Required
+
+- `node --test tests/commands/ai-analyze-project-provider.test.js`
+- `node --test tests/commands/ai-analyze-project.test.js`
+- `npm run docs:check`
+- `git diff --check`
+
+## Risks
+
+- Overly compact errors can hide necessary debugging detail.
+- Human output can accidentally break `--json`.
+
+## Dependencies
+
+- Depends on `slice-01-provider-fixture-harness`.
+
+## Instructions For Executor
+
+1. Preserve strict schema validation.
+2. Do not implement repair in this slice.
+3. Keep complete details in a manifest, not only in console output.
+
+## Completion Checklist
+
+- [ ] Error grouping implemented.
+- [ ] Tests assert compact output.
+- [ ] Manifest details are complete.
+- [ ] Closure brief updated.
+
+## Conditions Of Closure
+
+- Schema failure output is understandable without losing debug evidence.

package/specs/quiver-v53-reliable-deep-project-analysis/slices/slice-03-schema-error-grouping/slice.json

@@ -0,0 +1,80 @@
+{
+  "slice_id": "slice-03-schema-error-grouping",
+  "ticket": "QUIVER-53-03",
+  "type": "cli-ux",
+  "title": "Schema error grouping",
+  "objective": "Make provider schema failures compact, actionable, and fully auditable.",
+  "description": "Groups schema validation failures by type and path family in console output while writing complete validation details to a manifest for debugging.",
+  "git": {
+    "branch_type": "feature",
+    "base_branch": "main",
+    "branch_slug": "v53-schema-error-grouping",
+    "branch_name": "feature/QUIVER-53-03-v53-schema-error-grouping"
+  },
+  "files": [
+    "src/create-quiver/commands/ai.js",
+    "src/create-quiver/lib/ai/analyze-project-validation.js",
+    "src/create-quiver/lib/i18n/messages/en.js",
+    "src/create-quiver/lib/i18n/messages/es.js",
+    "tests/commands/ai-analyze-project-provider.test.js",
+    "specs/quiver-v53-reliable-deep-project-analysis/STATUS.md",
+    "specs/quiver-v53-reliable-deep-project-analysis/EVIDENCE_REPORT.md",
+    "specs/quiver-v53-reliable-deep-project-analysis/slices/slice-03-schema-error-grouping/CLOSURE_BRIEF.md",
+    "specs/quiver-v53-reliable-deep-project-analysis/slices/slice-03-schema-error-grouping/slice.json"
+  ],
+  "expected_read_paths": [
+    "src/create-quiver/commands/ai.js",
+    "src/create-quiver/lib/ai/analyze-project-validation.js",
+    "src/create-quiver/lib/ai/analyze-project-schema.js",
+    "tests/commands/ai-analyze-project-provider.test.js"
+  ],
+  "allowed_write_paths": [
+    "src/create-quiver/commands/ai.js",
+    "src/create-quiver/lib/ai/analyze-project-validation.js",
+    "src/create-quiver/lib/i18n/messages/en.js",
+    "src/create-quiver/lib/i18n/messages/es.js",
+    "tests/commands/ai-analyze-project-provider.test.js",
+    "tests/fixtures/analyze-project/**",
+    "specs/quiver-v53-reliable-deep-project-analysis/STATUS.md",
+    "specs/quiver-v53-reliable-deep-project-analysis/EVIDENCE_REPORT.md",
+    "specs/quiver-v53-reliable-deep-project-analysis/slices/slice-03-schema-error-grouping/CLOSURE_BRIEF.md",
+    "specs/quiver-v53-reliable-deep-project-analysis/slices/slice-03-schema-error-grouping/slice.json"
+  ],
+  "depends_on": [
+    "slice-01-provider-fixture-harness"
+  ],
+  "parallel_safe": "conditional",
+  "parallel_safe_reason": "Can run beside context-boundary work if command and validation files are coordinated.",
+  "must": [
+    "Group schema errors by issue type and path family.",
+    "Limit console examples to a small fixed number per failure.",
+    "Include path, type, cause hint, count, and next safe command.",
+    "Write complete validation issues to a manifest when provider execution creates a run."
+  ],
+  "not_included": [
+    "Repairing invalid JSON.",
+    "Retrying provider calls.",
+    "Changing the schema to accept extra keys."
+  ],
+  "acceptance": [
+    "The notes drift fixture produces compact console output instead of massive schema noise.",
+    "The complete validation issue list is available in a manifest.",
+    "The error message includes the next safe command.",
+    "--json output remains parseable."
+  ],
+  "tests": [
+    "node --test tests/commands/ai-analyze-project-provider.test.js",
+    "node --test tests/commands/ai-analyze-project.test.js",
+    "npm run docs:check",
+    "git diff --check"
+  ],
+  "validation_hints": [
+    "Assert an upper bound on console issue examples."
+  ],
+  "estimated_hours": 4,
+  "actual_hours": 3,
+  "status": "completed",
+  "started_at": "2026-06-11T12:03:20Z",
+  "completed_at": "2026-06-11T12:07:05Z",
+  "blocked_reason": null
+}

package/specs/quiver-v53-reliable-deep-project-analysis/slices/slice-04-safe-repair-layer/CLOSURE_BRIEF.md

@@ -0,0 +1,34 @@
+# CLOSURE_BRIEF - slice-04 safe repair layer
+
+## Status
+
+Completed on 2026-06-11.
+
+## Summary
+
+Safe repair now removes only allowlisted unsupported additional properties, currently `notes`, records every removed key, and revalidates the provider analysis before proceeding.
+
+## Evidence
+
+- Added `src/create-quiver/lib/ai/analyze-project-repair.js`.
+- Added repair unit tests under `tests/lib/ai/analyze-project-repair.test.js`.
+- Updated provider flow to use repair before failing schema validation.
+- Updated the `nika-erp` notes drift fixture expectation from fail-closed to repaired success.
+- Repair manifests are written to `.quiver/runs/run-.../repair/analyze-project-repair.json`.
+
+## Validation
+
+- `node --test tests/lib/ai/analyze-project-repair.test.js` passed: 3 tests.
+- `node --test tests/commands/ai-analyze-project-provider.test.js` passed: 14 tests.
+- `npm run schema:slice:check` passed.
+- `git diff --check` passed.
+
+## Decisions
+
+- Repair is intentionally narrow: only `unrecognized_keys` issues whose keys are in the safe allowlist are repaired.
+- `notes` content is removed, not moved or reinterpreted, to avoid semantic changes.
+- Unrepairable outputs still write validation evidence and fail without final docs writes.
+
+## Follow-ups
+
+- `slice-05-controlled-retry-layer` handles invalid outputs that cannot be repaired directly.

package/specs/quiver-v53-reliable-deep-project-analysis/slices/slice-04-safe-repair-layer/EXECUTION_BRIEF.md

@@ -0,0 +1,63 @@
+# EXECUTION_BRIEF - slice-04 safe repair layer
+
+## Context
+
+The observed drift is structurally repairable: unsupported `notes` keys appear in otherwise useful provider JSON. Repair must be safe, minimal, auditable, and non-semantic.
+
+## Objective
+
+Add safe repair for provider output that fails only because of unsupported additional properties.
+
+## Scope
+
+- Repair module.
+- Repair manifest.
+- Revalidation after repair.
+- Tests for `notes` drift and unrepairable errors.
+
+## Acceptance Criteria
+
+- Extra `notes` keys are removed and recorded.
+- Repaired output must pass schema before proceeding.
+- Repair does not move or reinterpret content.
+- Unrepairable errors still fail safely.
+
+## Expected Files To Modify
+
+- `src/create-quiver/lib/ai/analyze-project-repair.js`
+- `src/create-quiver/lib/ai/analyze-project-validation.js`
+- `src/create-quiver/commands/ai.js`
+- repair/provider tests
+
+## Validations Required
+
+- `node --test tests/lib/ai/analyze-project-repair.test.js`
+- `node --test tests/commands/ai-analyze-project-provider.test.js`
+- `npm run schema:slice:check`
+- `git diff --check`
+
+## Risks
+
+- Repair could hide real model mistakes if it becomes too broad.
+- Removing fields without manifest can reduce auditability.
+
+## Dependencies
+
+- Depends on `slice-03-schema-error-grouping`.
+
+## Instructions For Executor
+
+1. Keep repair whitelist narrow.
+2. Do not change semantic values.
+3. Record every repair and revalidate.
+
+## Completion Checklist
+
+- [ ] Repair module implemented.
+- [ ] Repair manifest implemented.
+- [ ] Revalidation tests pass.
+- [ ] Closure brief updated.
+
+## Conditions Of Closure
+
+- The `notes` drift failure can be repaired safely and audited.