Input from GPT-5.5 (Codex Desktop):

● ## PR Review: #10491

Status: Request Changes

Peer review of the .agent -> .agents migration. The target direction is reasonable, but the current branch is not merge-eligible because the diff contains unrelated work products and misses several required migration cleanups.

Context & Graph Linking

• Target Issue ID: Resolves #10489
• Related Graph Nodes: Progressive Disclosure Skills, harness startup context, ContextAssembler skill loading, Memory Core filesystem ingestion

Depth Floor

Challenge: The PR assumes the migration is complete because the primary skill tree moved, but residual .agent references and unrelated artifacts show the branch is not yet a clean canonical-directory migration.

Rhetorical-Drift Audit: Fails. The PR body says it “Deletes the deprecated .agent/ symlink hack and moves all skill directories natively into the plural .agents/ standard,” but the diff still has residual .agent references and includes unrelated generated files/scripts.

Graph Ingestion Notes

• [TOOLING_GAP]: The branch still contains generated SQLite DB files, WAL/SHM files, daemon logs, PID files, and scratch scripts. This is exactly the kind of local-test artifact contamination the review gate must catch.
• [RETROSPECTIVE]: Directory normalization is a legitimate cleanup, but it needs a narrow branch: canonical skill files and path references only.

Provenance Audit

N/A — this is a repository layout normalization, not a novel external abstraction.

Close-Target Audit

Fail. #10489 is not epic-labeled, so the target itself is valid, but the PR body uses prose-embedded close syntax: Resolves #10489. Deletes.... Required: isolate Resolves #10489 on its own line per the syntax-exact close-target mandate.

MCP-Tool-Description Budget Audit

Pass with note. The PR touches ai/mcp/server/github-workflow/openapi.yaml, but the modified descriptions are path-reference substitutions only; no new budget bloat observed.

Wire-Format Compatibility Audit

N/A — no wire-format payloads changed.

Cross-Skill Integration Audit

Fail. The migration touches skills, AGENTS.md, AGENTS_STARTUP.md, and runtime skill discovery, so all path consumers need to be audited. Current residual references include at least:

• .github/AI_QUICK_START.md:173 still points to <YOUR_NEO_REPO_PATH>/.agent/ANTIGRAVITY_RULES.md.
• ai/mcp/server/memory-core/services/FileSystemIngestor.mjs still ignores .agent but not .agents, which can make the canonical skill tree ingestible after the migration.
• AGENTS.md:23 still says “Throughout the .agent skill ecosystem”.

Required Actions

To proceed with merging, please address the following:

• Remove unrelated files from the PR diff: .codex/*, .gitignore, ai/scripts/summarize-sessions.mjs, package.json changes not required for #10489, generated SQLite DB/WAL/SHM files, test-daemon-dir/*, test-daemon-perf.mjs, and test-sqlite-insert.mjs, unless each is explicitly justified as part of #10489.
• Delete all generated local-test artifacts from the branch history/current diff: test-daemon-db.sqlite*, test-db.sqlite, daemon logs/PID files, and scratch test scripts.
• Fix remaining .agent path references that are actual directory-path references, especially .github/AI_QUICK_START.md:173, ai/mcp/server/memory-core/services/FileSystemIngestor.mjs, and AGENTS.md:23.
• Update the Memory Core file-ingestor ignore list so the new canonical .agents directory is ignored if skill content should remain excluded from filesystem ingestion.
• Fix git diff origin/dev...HEAD --check failures in AGENTS_STARTUP.md and ai/scripts/summarize-sessions.mjs, or remove the unrelated summarization file from this PR.
• Rewrite the PR body close target so Resolves #10489 is on its own line.

Evaluation Metrics

• [ARCH_ALIGNMENT]: 55 - 45 points deducted because the canonical-directory goal is valid, but residual path consumers and ingestion ignore rules are not fully migrated.
• [CONTENT_COMPLETENESS]: 45 - 55 points deducted because the PR body overstates completion and omits the unrelated artifacts currently present in the diff.
• [EXECUTION_QUALITY]: 35 - 65 points deducted because the branch includes generated binaries/logs/scratch scripts and fails git diff --check.
• [PRODUCTIVITY]: 50 - 50 points deducted because the primary migration is partially present, but the branch is too contaminated to safely merge.
• [IMPACT]: 60 - Substantive agent-workflow cleanup affecting all harnesses and skill discovery.
• [COMPLEXITY]: 65 - Medium-high: repository-wide path migration with runtime, docs, startup, and cross-harness blast radius.
• [EFFORT_PROFILE]: Heavy Lift - The file moves are mechanical, but correctness depends on exhaustive consumer cleanup and clean branch hygiene.

Verified

• gh pr view: state OPEN, base dev, merge state CLEAN, CodeQL success.
• git diff --name-status origin/dev...HEAD: shows unrelated added generated DB files/logs/scripts and unrelated Codex/summarization changes.
• rg path audit: found residual .agent directory references in .github/AI_QUICK_START.md, ai/mcp/server/memory-core/services/FileSystemIngestor.mjs, and AGENTS.md.
• git diff origin/dev...HEAD --check: fails on trailing whitespace.

Input from GPT-5.5 (Codex Desktop):

● ## PR Re-Review: #10491 head fe3669c

Status: Request Changes

I reviewed the force-pushed remote head after refreshing origin/dev. The generated SQLite scratch files are gone and git diff --check now passes, but the branch is still not merge-eligible.

Findings

1. [P1] Branch head/commit is wrong for this PR. The only commit on this PR head is fe3669ce5 chore(data): Hourly data sync pipeline update [skip ci]. That subject neither references #10489 nor describes the .agent -> .agents migration. The commit also carries [skip ci], and gh pr view reports an empty statusCheckRollup, so this PR currently has no CodeQL evidence on the reviewed head. Required: rebuild the branch with a proper chore(agents): ... (#10489) commit and without [skip ci] unless the branch truly is a data-sync pipeline exception.

2. [P1] Actual .agent path references remain. The migration still misses at least these real path references:

   • .github/AI_QUICK_START.md:173 still points to <YOUR_NEO_REPO_PATH>/.agent/ANTIGRAVITY_RULES.md.
   • AGENTS.md:23 still describes the “.agent skill ecosystem.” The rest of the rg hits are mostly class names like Neo.ai.agent.* and can stay, but these two path/prose references are part of the migration surface.

3. [P2] The PR still contains unrelated data-resource churn. After comparing against refreshed origin/dev, the diff still includes:

   • apps/devindex/resources/data/tracker.json
   • apps/devindex/resources/data/users.jsonl
   • apps/portal/sitemap.xml

   Those are not part of #10489’s directory-normalization scope. Required: remove them from this PR or explain why this agent-directory migration must carry data-sync churn.

4. [P2] .agents/skills/* symlink deletion should be intentional and documented. The diff deletes the symlink bridge entries introduced for Codex under .agents/skills/<skill> while renaming the actual .agent/skills/<skill>/... directories into .agents/skills/<skill>/.... That may be the right end state, but the PR body should explicitly call out that the symlink bridge is removed because the canonical directories now live at .agents/skills/*.

Close-Target Audit

Pass on syntax and target validity. Resolves #10489 is now on its own line, and #10489 is not epic-labeled.

MCP-Tool-Description Budget Audit

Pass with note. The OpenAPI changes are path-reference substitutions only, not description bloat.

Required Actions

To proceed with merging, please address the following:

• Rebuild the PR head with a correct conventional commit subject ending (#10489), not the current data-sync [skip ci] commit.
• Ensure CI/CodeQL runs on the reviewed head, or provide an explicit reason this PR should be exempt.
• Update the remaining real .agent references in .github/AI_QUICK_START.md:173 and AGENTS.md:23.
• Remove unrelated data-resource churn from apps/devindex/resources/data/tracker.json, apps/devindex/resources/data/users.jsonl, and apps/portal/sitemap.xml, unless intentionally justified in the PR body.
• Add one PR-body sentence documenting that the old .agents/skills/* symlink bridge is removed because .agents/skills/* now contains the canonical directories.

Verified

• Refreshed base: git fetch origin dev, now comparing against origin/dev at 7d7a99b29.
• git diff origin/dev...HEAD --check: passes.
• git diff --name-status origin/dev...HEAD: still includes the data-resource files listed above.
• rg path audit: real residual .agent references remain in .github/AI_QUICK_START.md and AGENTS.md.
• gh pr view: head fe3669ce5b7a5610045ffd9672ff3e7d98b287c7, merge state CLEAN, empty statusCheckRollup.

Input from GPT-5.5 (Codex Desktop):

● ## PR Re-Review: #10491 head 6e0d432

Status: Request Changes

The branch is now much closer: generated DB/scratch files and data-resource churn are gone, residual real .agent path references are fixed, CodeQL is green, and the PR body now documents removal of the symlink bridge. Two mechanical gates still block approval.

Findings

1. [P2] git diff --check still fails. AGENTS.md:23 has trailing whitespace on the changed .agents ecosystem sentence. Required: remove that trailing space and re-run git diff origin/dev...HEAD --check.

2. [P2] Commit subject is missing the required Conventional Commit scope. Current head commit is chore: migrate .agent to .agents directory (#10489). AGENTS.md §0/§3 requires type(scope): message (#NNNN). Required: amend to a scoped subject, e.g. chore(agents): migrate .agent to .agents directory (#10489).

Context & Graph Linking

• Target Issue ID: Resolves #10489
• Related Graph Nodes: Progressive Disclosure Skills, harness startup context, ContextAssembler skill loading, Memory Core filesystem ingestion

Depth Floor

Challenge: The core migration now appears scoped correctly, but this branch has repeatedly picked up mechanical artifacts. The remaining checks are exactly the branch-hygiene gates that keep those artifacts from reaching dev.

Rhetorical-Drift Audit: Pass. The PR body now matches the actual diff and explicitly explains that the old .agents/skills/* symlink bridge is removed because .agents/skills/* now hosts the canonical directories.

Graph Ingestion Notes

• [TOOLING_GAP]: The earlier data-sync contamination was removed after resetting against origin/dev; final approval is still blocked by whitespace and commit-format gates.
• [RETROSPECTIVE]: Directory migration reviews need both path-reference audits and branch-hygiene audits because file moves can obscure unrelated churn.

Provenance Audit

N/A — repository layout normalization, not a novel external abstraction.

Close-Target Audit

Pass. Resolves #10489 is newline-isolated, and #10489 is not epic-labeled.

MCP-Tool-Description Budget Audit

Pass. The OpenAPI changes are path-reference substitutions only; no description bloat observed.

Wire-Format Compatibility Audit

N/A — no wire-format payloads changed.

Cross-Skill Integration Audit

Pass with note. ContextAssembler now points at .agents/skills, startup/docs references are migrated, and FileSystemIngestor ignores .agents rather than .agent. rg no longer finds real stale .agent/ path references in the audited surfaces; remaining hits are class names/variable names such as Neo.ai.agent.*.

Required Actions

To proceed with merging, please address the following:

• Remove trailing whitespace at AGENTS.md:23 and confirm git diff origin/dev...HEAD --check passes.
• Amend the commit subject to include the required scope, e.g. chore(agents): migrate .agent to .agents directory (#10489).

Evaluation Metrics

• [ARCH_ALIGNMENT]: 95 - 5 points deducted only for the remaining commit-format gate; the directory standardization and runtime path changes now align with the intended .agents canonical layout.
• [CONTENT_COMPLETENESS]: 95 - 5 points deducted because the implementation prose is now accurate, but the branch still has one mechanical whitespace defect.
• [EXECUTION_QUALITY]: 85 - 15 points deducted because git diff --check fails and one targeted config spec fails with a namespace-collision setup error.
• [PRODUCTIVITY]: 90 - 10 points deducted because the migration scope is effectively complete, pending two small mechanical fixes.
• [IMPACT]: 60 - Substantive agent-workflow cleanup affecting all harnesses and skill discovery.
• [COMPLEXITY]: 60 - Medium: broad path migration across skill tree, docs, startup instructions, and runtime discovery/ingestion surfaces.
• [EFFORT_PROFILE]: Heavy Lift - Mostly mechanical file movement, but high blast radius across multiple agent harness surfaces.

Verified

• Refreshed remote state with git fetch origin dev chore/10489-migrate-agents-folder.
• Reviewed detached remote head 6e0d432c6661525ff484de7c73c5b083c3f344db.
• gh pr view: merge state CLEAN, CodeQL checks successful.
• git diff --name-status origin/dev...HEAD: no generated SQLite files, scratch scripts, or data-resource churn remain.
• rg path audit: no remaining real .agent/ path references in the audited surfaces; remaining matches are Neo.ai.agent class/member names or generic agent variables.
• Targeted tests: Assembler.spec.mjs passed both tests; ConfigCompleteness.spec.mjs failed during import with Namespace collision in unitTestMode for Neo.ai.mcp.server.github-workflow.Config, before reaching a path completeness assertion.

Input from GPT-5.5 (Codex Desktop):

● ## PR Re-Review: #10491 head 2549aee

Status: Approved

The latest head resolves the remaining mechanical blockers. The branch now has a scoped Conventional Commit subject, passes whitespace hygiene, has clean CodeQL checks, and the diff is scoped to the intended .agent -> .agents migration.

Context & Graph Linking

• Target Issue ID: Resolves #10489
• Related Graph Nodes: Progressive Disclosure Skills, harness startup context, ContextAssembler skill loading, Memory Core filesystem ingestion

Depth Floor

Documented search: I actively re-checked stale .agent/ path references, accidental data-resource churn, generated SQLite/test artifacts, symlink bridge leftovers, commit-message format, whitespace hygiene, OpenAPI description-budget drift, and runtime skill loading. No required concerns remain.

Rhetorical-Drift Audit: Pass. The PR body now matches the mechanical diff: the .agents/skills/* symlink bridge is removed because .agents/skills/* now contains the canonical skill directories.

Graph Ingestion Notes

• [TOOLING_GAP]: Earlier force-push/data-sync contamination was corrected; final head is clean and CodeQL-backed.
• [RETROSPECTIVE]: Directory migrations need both semantic path audits and file-scope audits; file moves can hide unrelated churn and stale runtime references.

Provenance Audit

N/A — repository layout normalization, not a novel external abstraction.

Close-Target Audit

Pass. Resolves #10489 is newline-isolated, and #10489 is not epic-labeled.

MCP-Tool-Description Budget Audit

Pass. The OpenAPI changes are path-reference substitutions only; no new description bloat observed.

Wire-Format Compatibility Audit

N/A — no wire-format payloads changed.

Cross-Skill Integration Audit

Pass. ContextAssembler now loads .agents/skills, startup/docs references are migrated, FileSystemIngestor ignores .agents, and no real stale .agent/ path references remain in the audited surfaces. Remaining rg hits are Neo.ai.agent.* class/member names or generic agent variables.

Required Actions

No required actions — eligible for human merge.

Evaluation Metrics

• [ARCH_ALIGNMENT]: 100 - I actively considered stale skill-loader paths, Memory Core ingestion paths, startup docs, and harness references and confirmed no required architecture gaps remain.
• [CONTENT_COMPLETENESS]: 95 - 5 points deducted only because the PR body is concise rather than a full Fat Ticket, but it now documents the symlink bridge removal and close target correctly.
• [EXECUTION_QUALITY]: 95 - 5 points deducted because ConfigCompleteness.spec.mjs still has the known unitTestMode namespace-collision blocker outside this PR; the directly relevant Assembler path test passes and CI is green.
• [PRODUCTIVITY]: 100 - Achieves #10489 acceptance criteria: canonical .agents layout, .agent removal, and path/config reference migration.
• [IMPACT]: 60 - Substantive agent-workflow cleanup affecting all harnesses and skill discovery.
• [COMPLEXITY]: 60 - Medium: broad path migration across skill tree, docs, startup instructions, OpenAPI descriptions, and runtime discovery/ingestion surfaces.
• [EFFORT_PROFILE]: Heavy Lift - Mostly mechanical file movement, but with high blast radius across multiple agent harness surfaces.

Verified

• Refreshed remote state with git fetch origin dev chore/10489-migrate-agents-folder.
• Reviewed detached remote head 2549aeefbe85ea7b2b21edf3ac0a627d3cf6608f.
• Commit subject: chore(agents): migrate .agent to .agents directory (#10489).
• git diff origin/dev...HEAD --check: passes.
• git diff --name-status origin/dev...HEAD: no generated SQLite files, scratch scripts, or data-resource churn remain.
• find .agents/skills -maxdepth 1 -type l -print: no symlinks remain in .agents/skills.
• rg path audit: no remaining real .agent/ path references in audited surfaces.
• npm run test -- test/playwright/unit/ai/context/Assembler.spec.mjs: 2 passed (704ms).
• gh pr view: merge state CLEAN, CodeQL checks successful.

Input from GPT-5.5 (Codex Desktop):

● ## PR Re-Review Correction: #10491 head 2549aee

Status: Request Changes

Supersedes my previous approval. @tobiu caught a merge-gate issue that my review missed: the migration removes .agent, but leaves the Claude skill symlink bridge pointing at .agent.

Finding

[P1] Claude skill symlinks break after .agent is removed.

Empirical check on head 2549aeefbe85ea7b2b21edf3ac0a627d3cf6608f:

• .claude/skills is byte-for-byte unchanged from origin/dev.
• Example symlink target: .claude/skills/pr-review -> ../../.agent/skills/pr-review.
• .agent/skills/pr-review does not exist on this PR head.
• .agents/skills/pr-review does exist.
• Full audit: all 16 tracked .claude/skills/* symlinks point to ../../.agent/skills/* and are broken on this head.

This means Claude would lose its skill discovery after the migration, even though Codex/Antigravity paths are updated.

Required Action

To proceed with merging, please address the following:

• Update all .claude/skills/* symlinks to point at ../../.agents/skills/<skill> instead of ../../.agent/skills/<skill>, or otherwise preserve Claude skill resolution under the new canonical .agents layout.
• Re-run a symlink-resolution audit confirming every .claude/skills/* target exists on the PR head.

Verification Commands Used

• git show origin/chore/10489-migrate-agents-folder:.claude/skills/pr-review returned ../../.agent/skills/pr-review.
• git cat-file -e origin/chore/10489-migrate-agents-folder:.agent/skills/pr-review failed.
• git cat-file -e origin/chore/10489-migrate-agents-folder:.agents/skills/pr-review passed.

No approval can stand while Claude’s skill symlinks are broken.

Input from GPT-5.5 (Codex Desktop):

● ## PR Re-Review: #10491 head 1464392

Status: Request Changes

The Claude symlink blocker is fixed: all 16 .claude/skills/* symlinks now point to ../../.agents/skills/<skill> and resolve on the PR head. However, the broader post-merge path audit surfaced one remaining P1 blocker: the moved skill files still contain active .agent/skills/... router instructions even though this PR removes .agent from the tree.

Context & Graph Linking

• Target Issue ID: Resolves #10489
• Related Graph Nodes: Progressive Disclosure Skills, harness startup context, Claude Code skill discovery, ContextAssembler skill loading, Memory Core filesystem ingestion

Depth Floor

Challenge: The branch now fixes the visible Claude symlink bridge, but it still misses the skill files as downstream consumers of the path migration. A skill can be discovered from .agents/skills, then instruct the agent to read a nonexistent .agent/skills/... reference file.

Rhetorical-Drift Audit: Fails narrowly. The PR body says internal reference pointers are migrated to .agents, but active router text inside the moved skill files still points at .agent/skills/....

Finding

1. [P1] Moved skill files still route agents to deleted .agent/skills paths. Examples on head 14643920f65a148a1c531e833d809c7e7eae3689:

   • .agents/skills/pr-review/SKILL.md:8 tells reviewers to read .agent/skills/pr-review/references/pr-review-guide.md.
   • .agents/skills/pull-request/SKILL.md:8 tells PR authors to read .agent/skills/pull-request/references/pull-request-workflow.md.
   • .agents/skills/ticket-intake/SKILL.md:11 tells agents to read .agent/skills/ticket-intake/references/ticket-intake-workflow.md.
   • .agents/skills/create-skill/references/skill-authoring-guide.md:79 still shows ln -sf ../../.agent/skills/my-new-skill .claude/skills/my-new-skill.

   git ls-tree origin/chore/10489-migrate-agents-folder .agent returns no .agent tree, so these are dead-path instructions after merge.

Graph Ingestion Notes

• [TOOLING_GAP]: The first local Playwright run hit sandbox EPERM binding 0.0.0.0:8080; rerun with approved escalation passed. This is environment friction, not a PR defect.
• [RETROSPECTIVE]: Directory migrations must audit both mechanical symlinks and instructional path literals inside the moved assets. The latter can break agent behavior even when runtime tests and symlink checks pass.

Provenance Audit

N/A — repository layout normalization, not a new external abstraction.

Close-Target Audit

Pass. Resolves #10489 is newline-isolated, and #10489 carries ai, refactoring, and architecture, not epic.

MCP-Tool-Description Budget Audit

Pass with note. The OpenAPI changes are path-reference substitutions only; no new tool-description narrative or budget bloat observed.

Wire-Format Compatibility Audit

N/A — no wire-format payloads changed.

Cross-Skill Integration Audit

Request changes. Skill discovery paths now resolve through .agents, but the skill bodies still contain active .agent/skills follow-up paths and symlink-authoring examples.

Required Actions

To proceed with merging, please address the following:

• Update active .agent/skills/... references inside .agents/skills/** to .agents/skills/..., including top-level SKILL.md router instructions, reference-file links, and create-skill symlink examples.
• Re-run a path-literal audit such as git grep -n -F ".agent/" origin/chore/10489-migrate-agents-folder -- .agents/skills .github AGENTS.md AGENTS_STARTUP.md README.md ROADMAP.md ai learn .claude; remaining hits should be either gone or explicitly historical/non-operative.

Evaluation Metrics

• [ARCH_ALIGNMENT]: 85 - 15 points deducted because the canonical .agents direction is correct, but active skill-router paths still point at the removed canonical directory.
• [CONTENT_COMPLETENESS]: 80 - 20 points deducted because the PR body and moved skill documentation do not yet fully reflect the post-migration path reality.
• [EXECUTION_QUALITY]: 80 - 20 points deducted because whitespace, CI, tests, and symlinks are clean, but a P1 post-merge behavioral break remains for agents following skill instructions.
• [PRODUCTIVITY]: 85 - 15 points deducted because the primary migration is nearly complete, but the active skill-instruction surface still blocks #10489 completion.
• [IMPACT]: 60 - Substantive agent-workflow cleanup affecting all harnesses and skill discovery.
• [COMPLEXITY]: 60 - Medium: broad path migration across skill tree, startup docs, runtime discovery, Memory Core ingestion, and harness-specific skill bridges.
• [EFFORT_PROFILE]: Heavy Lift - Mostly mechanical, but high blast radius across agent lifecycle surfaces.

Verified

• Refreshed remote state with git fetch origin dev chore/10489-migrate-agents-folder.
• Reviewed remote head 14643920f65a148a1c531e833d809c7e7eae3689 against current origin/dev at 7e1bbc03bf745715e322bd243dd53360d242fa1a.
• Commit subject: chore(agents): migrate .agent to .agents directory (#10489).
• git diff origin/dev...origin/chore/10489-migrate-agents-folder --check: passes.
• gh pr view: merge state CLEAN; CodeQL checks successful.
• .claude/skills symlink audit: AUDIT PASS: 16 .claude/skills symlinks checked.
• npm run test -- test/playwright/unit/ai/context/Assembler.spec.mjs: 2 passed (825ms) after sandbox escalation.
• git grep -n -F ".agent/" ... -- .agents/skills ...: still finds active .agent/skills path literals inside moved skill files.

Input from GPT-5.5 (Codex Desktop):

● ## PR Re-Review: #10491 head 5851ab7

Status: Approved

The latest head resolves the remaining blockers. The Claude skill bridge now resolves under .agents, the moved skill router instructions no longer point at deleted .agent/skills/... paths, and the branch remains scoped to the intended .agent -> .agents migration.

Context & Graph Linking

• Target Issue ID: Resolves #10489
• Related Graph Nodes: Progressive Disclosure Skills, harness startup context, Claude Code skill discovery, ContextAssembler skill loading, Memory Core filesystem ingestion

Depth Floor

Documented search: I actively re-checked active .agent/ path literals across .agents/skills, startup docs, runtime docs, AI server files, and .claude; .claude/skills symlink resolution; accidental generated/data-resource churn; whitespace hygiene; commit-message format; close-target labels; OpenAPI description-budget drift; CodeQL status; and the targeted Assembler skill-loading test. No required concerns remain.

Rhetorical-Drift Audit: Pass. The PR body says internal reference pointers are migrated to .agents, and the current path-literal audit confirms no active .agent/ references remain in the audited surfaces.

Graph Ingestion Notes

• [TOOLING_GAP]: Local Playwright verification requires sandbox escalation to let the test web server bind its port in this harness. This is environment friction, not a PR defect.
• [RETROSPECTIVE]: The Tobi merge-gate caught the cross-harness Claude symlink gap; the follow-up review expanded the migration checklist to include internal skill-router literals, which is the right audit shape for future directory migrations.

Provenance Audit

N/A — repository layout normalization, not a new external abstraction.

Close-Target Audit

Pass. Resolves #10489 is newline-isolated, and #10489 carries ai, refactoring, and architecture, not epic.

MCP-Tool-Description Budget Audit

Pass. The OpenAPI changes are path-reference substitutions only; no new tool-description narrative or budget bloat observed.

Wire-Format Compatibility Audit

N/A — no wire-format payloads changed.

Cross-Skill Integration Audit

Pass. ContextAssembler, startup docs, Memory Core ingestion ignores, .claude/skills symlinks, and moved skill router/reference paths now all point at the canonical .agents layout.

Required Actions

No required actions — eligible for human merge.

Evaluation Metrics

• [ARCH_ALIGNMENT]: 100 - I actively considered stale skill-loader paths, harness-specific skill bridges, Memory Core ingestion paths, startup docs, and moved skill-router instructions and confirmed no required architecture gaps remain.
• [CONTENT_COMPLETENESS]: 95 - 5 points deducted only because the PR body remains concise rather than a full Fat Ticket; the implementation/docs now accurately describe the canonical .agents layout.
• [EXECUTION_QUALITY]: 95 - 5 points deducted only because verification is targeted rather than a full suite run; diff hygiene, CodeQL, symlink audit, path-literal audit, and Assembler test all pass.
• [PRODUCTIVITY]: 100 - Achieves #10489 acceptance criteria: canonical .agents layout, .agent removal, and cross-harness path/config reference migration.
• [IMPACT]: 60 - Substantive agent-workflow cleanup affecting all harnesses and skill discovery.
• [COMPLEXITY]: 60 - Medium: broad path migration across skill tree, startup docs, runtime discovery, Memory Core ingestion, and harness-specific skill bridges.
• [EFFORT_PROFILE]: Heavy Lift - Mostly mechanical, but high blast radius across agent lifecycle surfaces.

Verified

• Refreshed remote state with git fetch origin dev chore/10489-migrate-agents-folder.
• Reviewed remote head 5851ab70a540687787425252276c5da0e7b43b23 against current origin/dev at 7e1bbc03bf745715e322bd243dd53360d242fa1a.
• Commit subject: chore(agents): migrate .agent to .agents directory (#10489).
• git diff --check origin/dev...origin/chore/10489-migrate-agents-folder: passes.
• git grep -n -F ".agent/" origin/chore/10489-migrate-agents-folder -- .agents/skills .github AGENTS.md AGENTS_STARTUP.md README.md ROADMAP.md ai learn .claude: no results.
• git ls-tree origin/chore/10489-migrate-agents-folder .agent: no .agent tree remains.
• .claude/skills symlink audit: AUDIT PASS: 16 .claude/skills symlinks checked.
• git diff --name-status origin/dev...origin/chore/10489-migrate-agents-folder: scoped to the .agent -> .agents migration and related path references; no generated SQLite files, scratch scripts, or data-resource churn.
• npm run test -- test/playwright/unit/ai/context/Assembler.spec.mjs: 2 passed (705ms).
• gh pr view: merge state CLEAN; CodeQL checks successful.