oat-project-review-provide

name	oat-project-review-provide
version	1.3.10
description	Use when the user explicitly asks to review an OAT project — e.g. "review project", "review the project", "run project review", or confirms a previously offered review. Do NOT auto-invoke on completed work alone. Resolves a project review scope and offers before running.
disable-model-invocation	false
user-invocable	true
allowed-tools	Read, Glob, Grep, Write, Edit, Bash(git:*), Bash(oat:*), Bash(pnpm:*), Bash(mkdir:*), Bash(date:*), Bash(realpath:*), Bash(awk:*), AskUserQuestion

Request Review

Request and execute a code or artifact review for the current project scope.

Purpose

Produce an independent review artifact that verifies requirements/design alignment (mode-aware) and code quality.

Reviewers should distinguish implementation defects from artifact drift. If code is defensible but spec.md, design.md, or plan.md is stale, frame the finding as artifact alignment rather than a required code change.

Prerequisites

Required: Active project or explicit user-provided project/review target that resolves to project state, with at least one completed task.

Required: Core project artifacts are already committed before the review begins. Review should not be the first step that notices an untracked project tree or pending bookkeeping-only artifact edits.

Model Invocation Gate

This skill is model-invokable only for explicit review asks such as "review project" or "review the project", or when the user confirms a previously offered project-review step. A gate-originated request from oat gate review is also an explicit review ask. Do NOT auto-invoke merely because a task, phase, or implementation appears complete.

Before acting, verify that there is an active OAT project or a user-provided review target that can be resolved to project state. If neither exists, do not run this skill; offer oat-project-open / oat-project-quick-start for project workflow setup, or oat-review-provide for a non-project ad-hoc review.

When the invocation is manual, summarize the inferred review type and scope, then ask before running the review.

Gate-originated mode: If the request context says the review is gate-originated, set REVIEW_INVOCATION=gate, honor any provided review type/scope arguments, and run without interactive confirmation prompts. The gate session is already the explicit authorization boundary. If subagent dispatch is unavailable, run the review inline in the current gate session instead of handing off to a fresh session.

Mode Assertion

OAT MODE: Review Request

Purpose: Determine review scope and execute a fresh-context review.

Progress Indicators (User-Facing)

When executing this skill, provide lightweight progress feedback so the user can tell what’s happening after they confirm.

• Print a phase banner once at start using horizontal separators, e.g.:

  ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ OAT ▸ PROVIDE REVIEW ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

• Before multi-step work (scope resolution, file gathering, writing artifact), print 2–5 short step indicators, e.g.:

  • [1/5] Resolving scope + range…
  • [2/5] Collecting files + context…
  • [3/5] Checking subagent availability…
  • [4/5] Running review…
  • [5/5] Writing review artifact…

• For long-running operations (reviewing large diffs, running verification commands), print a start line and a completion line (duration optional).

• Keep it concise; don’t print a line for every shell command.

BLOCKED Activities:

• No code changes during review
• No fixing issues found (that comes in receive-review)

ALLOWED Activities:

• Reading artifacts and code
• Running verification commands
• Writing review artifact

Usage

With arguments (if supported)

oat-project-review-provide code p02          # Code review for phase
oat-project-review-provide code p02-p03      # Code review for contiguous phase range
oat-project-review-provide code p02-t03      # Code review for task
oat-project-review-provide code final        # Final code review
oat-project-review-provide code base_sha=abc # Review since specific SHA
oat-project-review-provide artifact discovery # Artifact review of discovery.md
oat-project-review-provide artifact spec     # Artifact review of spec.md
oat-project-review-provide artifact design   # Artifact review of design.md

Without arguments

Run the oat-project-review-provide skill and it will:

1. Ask review type (code or artifact)
2. Ask scope (task/phase/final/range)
3. Confirm before running

Process

Step 0: Resolve Project or Explicit Review Target

OAT stores active project context in .oat/config.local.json (activeProject, local-only).

PROJECT_PATH=$(oat config get activeProject 2>/dev/null || true)
PROJECTS_ROOT="${OAT_PROJECTS_ROOT:-$(oat config get projects.root 2>/dev/null || echo ".oat/projects/shared")}"
PROJECTS_ROOT="${PROJECTS_ROOT%/}"

Validation rules:

• Prefer PROJECT_PATH from activeProject when it is set and points to an existing directory.
• If activeProject is missing or invalid, allow an explicit user-provided project/review target to resolve PROJECT_PATH:
  • project path, e.g. .oat/projects/shared/{project-name}
  • project name, resolved as ${PROJECTS_ROOT}/{project-name}
  • review target phrasing that includes a project name or project path
• A resolved PROJECT_PATH must point to an existing directory.
• "$PROJECT_PATH/state.md" must exist for mode-aware project review validation.

If neither an active project nor an explicit target resolves to a valid PROJECT_PATH with state.md, stop and route. Do not create or guess project pointers in this skill.

Tell user:

• This is a project-scoped skill and needs an initialized OAT project, either from activeProject or from a project/review target the user explicitly provided.
• Without resolvable project state, review can still proceed via non-project skill: oat-review-provide.
• To continue with project workflow instead, run one of:
  • oat-project-open (existing project)
  • oat-project-quick-start (new quick project)
  • oat-project-import-plan (external plan import)

If validation passes, derive {project-name} as basename of PROJECT_PATH. Summarize the resolved project/review target. For manual invocation, ask before continuing to Step 1. For REVIEW_INVOCATION=gate, continue without asking.

Step 1: Parse Arguments or Ask

If arguments provided:

• Parse $ARGUMENTS[0] as review type: code or artifact
• Parse $ARGUMENTS[1] as scope token

If no arguments — infer from project state:

Read state.md frontmatter to propose the most likely review type and scope:

eval "$(oat project status --shell \
  PHASE=project.phase \
  PHASE_STATUS=project.phaseStatus \
  WORKFLOW_MODE=project.workflowMode 2>/dev/null)"

Inference rules (first match wins):

Phase	Status	Inferred review
discovery	complete	artifact discovery
spec	complete	artifact spec
design	complete	artifact design
plan	complete	artifact plan
implement	in_progress	code with current phase scope (derive from implementation.md current task)
implement	complete	code final

If inference produces a result, propose it and proceed unless the user overrides:

Based on project state ({phase}, {phase_status}), I'd run: {type} {scope}
Proceed? (Y / or specify different type scope)

If the user confirms (or just presses enter), use the inferred type and scope. If the user provides an alternative, use that instead.

If state.md is missing or phase is unrecognized: fall back to asking:

• Ask: "What type of review? (code / artifact)"
• Ask: "What scope?"
  • For code: pNN-tNN task / pNN phase / pNN-pMM contiguous phase range / final / base_sha=SHA / SHA..HEAD range
  • For artifact: discovery / spec / design (and optionally plan)

Step 1.5: Resolve Target Branch and Working Directory

Before validating artifacts or gathering files, verify that the review will run and write artifacts against the correct branch and working directory. This step must run before Step 2 so that PROJECT_PATH points to the correct checkout for all downstream operations.

Detect current state:

CURRENT_BRANCH=$(git branch --show-current)

Handle detached HEAD: If CURRENT_BRANCH is empty (detached HEAD), resolve the branch from the current worktree entry:

if [[ -z "$CURRENT_BRANCH" ]]; then
  REPO_ROOT=$(git rev-parse --show-toplevel)
  CURRENT_BRANCH=$(git worktree list --porcelain | awk -v wt="$REPO_ROOT" '
    /^worktree / { cur=$2 }
    /^branch / { if (cur == wt) { sub("refs/heads/", "", $2); print $2 } }
  ')
fi

If still empty after worktree lookup, ask user: "Unable to detect current branch (detached HEAD). Which branch should the review target?"

TARGET_BRANCH="${TARGET_BRANCH:-$CURRENT_BRANCH}"  # from user scope, worktree context, or default

If target branch matches current branch: proceed normally — no path adjustment needed.

If target branch differs from current branch:

1. Check if the target branch has a worktree:

   WORKTREE_PATH=$(git worktree list --porcelain | awk -v branch="$TARGET_BRANCH" '
     /^worktree / { wt=$2 }
     /^branch / { if ($2 == "refs/heads/" branch) print wt }
   ')
   

2. If worktree exists for target branch:

   • Compute the project's relative path within the repo: REL_PROJECT=$(realpath --relative-to="$(git rev-parse --show-toplevel)" "$PROJECT_PATH")
   • Update PROJECT_PATH to resolve inside the worktree: PROJECT_PATH="$WORKTREE_PATH/$REL_PROJECT"
   • All subsequent steps (artifact validation, git diff/log, artifact writes, commits) use the updated PROJECT_PATH.
   • Run git commands scoped to the worktree: git -C "$WORKTREE_PATH" ...
   • Print: Review target: worktree at {WORKTREE_PATH} (branch: {TARGET_BRANCH})

3. If no worktree exists (regular branch on main worktree):

   • Stop and notify the user. Do not silently write to the wrong branch.

   • Print:

     ⚠️  Target branch "{TARGET_BRANCH}" differs from current branch "{CURRENT_BRANCH}".
     No worktree found for "{TARGET_BRANCH}".
     
     Options:
     1. Switch to branch "{TARGET_BRANCH}" (git checkout) — artifact will be written on that branch
     2. Provide review inline only (no artifact written to disk)
     3. Cancel and create a worktree first
     
     Choose:
     

   • If user chooses option 1: run git checkout {TARGET_BRANCH}, then proceed.

   • If user chooses option 2: set INLINE_ONLY=true — skip artifact write (Step 7/8) and output review findings directly in the session. The user can manually save the output.

Step 1.6: Enforce Committed Artifact Baseline

Before gathering review context, inspect the core project artifacts:

• "$PROJECT_PATH/discovery.md"
• "$PROJECT_PATH/spec.md"
• "$PROJECT_PATH/design.md"
• "$PROJECT_PATH/plan.md"
• "$PROJECT_PATH/implementation.md"
• "$PROJECT_PATH/state.md"
• .oat/state.md is generated dashboard state; ignore it for committed artifact baseline checks.

If any of those files are untracked or modified only because the previous workflow step did not finish its bookkeeping commit:

• Stop and tell the user to commit the pending artifact bookkeeping first, or resume the originating workflow skill so it can do that commit.
• Do not write a review artifact against that half-tracked state.

If the review is intentionally inline-only and the user explicitly wants to inspect an uncommitted artifact state, say so clearly in the output and skip writing the review artifact to disk.

• If user chooses option 3: stop and suggest oat-worktree-bootstrap-auto.

Step 2: Validate Artifacts Exist (Mode-Aware)

Resolve workflow mode from the resolved project state path:

# Step 1.5 may retarget PROJECT_PATH into another worktree, so read from the
# resolved project path instead of the active-project pointer.
WORKFLOW_MODE=$(oat project status --project-path "$PROJECT_PATH" --field project.workflowMode 2>/dev/null || echo null)

Required for code review (by mode):

• spec-driven: spec.md, design.md, plan.md
• quick: discovery.md, plan.md (spec.md/design.md optional if present)
• import: plan.md (references/imported-plan.md recommended, spec.md/design.md optional)

Required for artifact review:

• The artifact being reviewed must exist.
• Upstream dependencies are required only when relevant to that artifact:
  • reviewing spec requires discovery.md
  • reviewing design in spec-driven mode requires spec.md
  • reviewing design in quick/import mode requires only discovery.md (spec is skipped in these modes)
  • in quick/import mode, missing spec.md must not be treated as a project review gate failure for artifact design; proceed with normal project-scoped review flow, artifact writing, and bookkeeping
  • reviewing plan in spec-driven mode requires spec.md + design.md
  • reviewing plan in quick/import mode may use discovery.md and/or references/imported-plan.md instead

If missing: Report missing required artifacts for the current mode and stop if requirements are not met.

Step 3: Determine Scope and Commits

If review type is artifact:

• Interpret the scope token as the artifact name (discovery, spec, design, or plan)
• Set SCOPE_RANGE="" (no git range required)
• Proceed to Step 5 (metadata); Step 4 uses artifact files, not git diff

If review type is code, use the scope resolution below.

Step 3a: Detect Re-Review Context

Before resolving scope, check if this is a re-review of fixes from a prior review cycle:

1. Scan plan.md for tasks tagged with (review) in the scope being reviewed (e.g., (p02-review) fix tasks for a p02 phase review or (p02-p03-review) for a contiguous phase-range review).

2. If (review) fix tasks exist and their status is completed:

   • This is a re-review. Before prompting, check the workflow preference:

     AUTO_NARROW=$(oat config get workflow.autoNarrowReReviewScope 2>/dev/null || true)
     

     • If AUTO_NARROW is true: Auto-narrow. Print Re-review scope: narrowed to fix commits (from workflow.autoNarrowReReviewScope). Gather only the commits for completed (review) fix tasks (see below). Skip the prompt.
     • If AUTO_NARROW is false: Use full scope. Print Re-review scope: full (from workflow.autoNarrowReReviewScope). Skip the prompt and proceed with full scope resolution below.
     • If unset: Fall through to the standard prompt.

   • Standard prompt (when preference is unset):

     Detected completed review fix tasks for this scope:
     - {task IDs and descriptions}
     
     Scope to fix task commits only? (Y/n)
     

   • If yes (default): gather only the commits associated with those fix tasks using commit convention grep (e.g., git log --oneline --grep="\(pNN-tNN\)" HEAD~50..HEAD for each fix task ID). Set SCOPE_RANGE to cover only those commits.

   • If no: proceed with full scope resolution below (re-review everything).

3. If no (review) fix tasks exist, or they are not yet completed, proceed with normal scope resolution.

Priority order for scope resolution:

1. Explicit user input (preferred):
   • base_sha=<sha> → review range is <sha>..HEAD
   • <sha1>..<sha2> → exact range review
   • pNN-tNN → task scope
   • pNN → phase scope
   • pNN-pMM → contiguous inclusive phase-range scope (for example p02-p03)
   • final → full project review

Phase-range semantics:

• pNN-pMM scopes are inclusive and must represent contiguous implementation phases.
• This is the canonical scope format for checkpoint auto-reviews that need to cover multiple previously unpassed phases in one review.
• When a phase-range token is used, treat it as a range review for artifact naming/storage, but preserve the exact oat_review_scope value (for example p02-p03) in frontmatter and plan review rows.

2. Automatic phase detection (if invoked at phase boundary):

   • Derive current phase from plan.md + implementation.md

   • Use commit convention grep to find commits:

     # Task commits: grep for (pNN-tNN)
     git log --oneline --grep="\(p${PHASE}-t" HEAD~50..HEAD
     
     # Phase commits: grep for (pNN-
     git log --oneline --grep="\(p${PHASE}-" HEAD~50..HEAD
     

   • For contiguous phase-range scopes (pNN-pMM), aggregate commit matches for each phase in the inclusive range:

     for PHASE_NUM in $(seq "$START_PHASE_NUM" "$END_PHASE_NUM"); do
       PHASE_ID=$(printf "p%02d" "$PHASE_NUM")
       git log --oneline --grep="\\(${PHASE_ID}-" HEAD~50..HEAD
     done
     

3. Fallback (if commit conventions missing/inconsistent):

   • Prompt user to choose:
     • Provide base_sha=<sha>
     • Provide <sha1>..<sha2> range
     • Confirm "review merge-base..HEAD" (all changes on branch)

   Merge-base approach:

   MERGE_BASE=$(git merge-base origin/main HEAD 2>/dev/null || git merge-base main HEAD 2>/dev/null)
   SCOPE_RANGE="$MERGE_BASE..HEAD"
   

Step 4: Get Files Changed

If review type is code, once scope range is determined:

FILES_CHANGED=$(git diff --name-only "$SCOPE_RANGE" 2>/dev/null)
FILE_COUNT=$(echo "$FILES_CHANGED" | wc -l | tr -d ' ')

If review type is artifact, the "files in scope" are the artifact(s):

case "$SCOPE_TOKEN" in
  discovery) FILES_CHANGED=$(printf "%s\n" "$PROJECT_PATH/discovery.md") ;;
  spec) FILES_CHANGED=$(printf "%s\n" "$PROJECT_PATH/spec.md" "$PROJECT_PATH/discovery.md") ;;
  design)
    if [[ "$WORKFLOW_MODE" == "spec-driven" ]]; then
      FILES_CHANGED=$(printf "%s\n" "$PROJECT_PATH/design.md" "$PROJECT_PATH/spec.md")
    else
      FILES_CHANGED=$(printf "%s\n" "$PROJECT_PATH/design.md" "$PROJECT_PATH/discovery.md")
    fi
    ;;
  plan)
    if [[ "$WORKFLOW_MODE" == "spec-driven" ]]; then
      FILES_CHANGED=$(printf "%s\n" "$PROJECT_PATH/plan.md" "$PROJECT_PATH/spec.md" "$PROJECT_PATH/design.md")
    elif [[ "$WORKFLOW_MODE" == "quick" ]]; then
      FILES_CHANGED=$(printf "%s\n" "$PROJECT_PATH/plan.md" "$PROJECT_PATH/discovery.md")
    else
      FILES_CHANGED=$(printf "%s\n" "$PROJECT_PATH/plan.md" "$PROJECT_PATH/references/imported-plan.md")
    fi
    ;;
esac
FILE_COUNT=$(echo "$FILES_CHANGED" | wc -l | tr -d ' ')

Display to user:

Review scope: {scope}
Range: {SCOPE_RANGE} (code reviews only; artifact reviews have no git range)
Files changed: {FILE_COUNT}

{FILE_LIST preview - first 20 files}

Proceed with review?

If REVIEW_INVOCATION=gate, print the same scope summary but do not ask Proceed with review?; continue directly.

Step 4.1: Dispatch Profile Override Advisory (Artifact Plan Only)

When reviewing artifact plan, apply this Dispatch Profile override advisory:

• A missing ## Dispatch Profile section is normal and must not be flagged.
• Important findings:
  • invalid phase ID that does not match a real plan phase
  • unknown active-provider tier value
  • low-tier override for multi-file integration, architecture, or review-heavy work
  • low-tier override with missing or generic rationale
• Medium findings:
  • malformed but recoverable Dispatch Profile table structure
  • mid-tier override for architecture-heavy work without convincing rationale
• Minor findings:
  • rationale is present but weakly tied to phase scope

Include this advisory in the Review Scope metadata for artifact plan reviews so the reviewer evaluates explicit override rows without treating omitted rows as a gap.

Step 4.5: Gather Deferred Findings Ledger (Final Scope Only)

If review type == code and scope == final, gather unresolved deferred findings from prior review cycles.

Preferred sources:

• implementation.md sections titled Deferred Findings (...)
• prior review artifacts under reviews/archived/ when implementation notes are incomplete (plus the current active review file in reviews/, if one exists for the in-flight cycle)

Build:

• DEFERRED_MEDIUM_COUNT
• DEFERRED_MINOR_COUNT
• DEFERRED_LEDGER (one-line summary per finding with source artifact)

Rules:

• Include this ledger in review metadata so final review explicitly re-evaluates carry-forward debt.
• Final review should call out whether each deferred Medium remains acceptable or should now be fixed.

Step 5: Prepare Review Metadata Block

Build the "Review Scope" metadata for the reviewer:

## Review Scope

**Project:** {PROJECT_PATH}
**Type:** {code|artifact}
**Scope:** {scope}{optional: " (" + SCOPE_RANGE + ")"}
**Date:** {today}

**Artifact Paths:**

- Spec: {PROJECT_PATH}/spec.md (required in spec-driven mode; optional in quick/import)
- Design: {PROJECT_PATH}/design.md (required in spec-driven mode; optional in quick/import)
- Plan: {PROJECT_PATH}/plan.md
- Implementation: {PROJECT_PATH}/implementation.md
- Discovery: {PROJECT_PATH}/discovery.md
- Imported Plan Reference: {PROJECT_PATH}/references/imported-plan.md (optional; import mode)

**Tasks in Scope (code review only):** {task IDs from plan.md matching scope}

**Files Changed ({FILE_COUNT}):**
{FILE_LIST}

**Commits (code review only):**
{git log --oneline for SCOPE_RANGE}

**Deferred Findings Ledger (final scope only):**

- Deferred Medium count: {DEFERRED_MEDIUM_COUNT}
- Deferred Minor count: {DEFERRED_MINOR_COUNT}
  {DEFERRED_LEDGER}

**Design Drift Review Guidance:**

- If implementation differs from `spec.md`, `design.md`, or `plan.md`, decide whether the code should change or whether the artifact is stale.
- Use artifact-alignment framing when shipped implementation is defensible and the lifecycle artifact should be updated.
- Do not force a code-defect framing for accepted design drift; `oat-project-review-receive` can convert artifact drift into alignment tasks or explicit deferrals.

Step 6: Execute Review (3-Tier Capability Model)

Step 6a: Probe Subagent Availability

Before selecting a tier, announce the probe and its result so the user can see what's happening:

[3/5] Checking subagent availability…
  → oat-reviewer: {available | authorization required | not resolved} ({reason})
  → Selected: Tier {1|2|3} — {Subagent (fresh context) | Fresh session (recommended) | Inline review}

Detection logic:

• If the host is Claude Code, use Task-style subagent dispatch with subagent_type: "oat-reviewer" and resolve from .claude/agents/oat-reviewer.md.

• If the host is Cursor, invoke oat-reviewer using Cursor-native explicit invocation (/oat-reviewer) or natural mention, and resolve from .cursor/agents/oat-reviewer.md (or .claude/agents/oat-reviewer.md compatibility path).

• If the host is Codex multi-agent, verify Codex requirements first:

  • [features] multi_agent = true is enabled in active Codex config.

  • If explicit role pinning is desired, agent_type must be a built-in role (default/worker/explorer) or a custom role declared under [agents.<name>].

  • Codex may also auto-select and spawn agents without explicit role pinning.

  • If the current Codex host requires explicit user authorization before calling spawn_agent, do not mark oat-reviewer as unresolved. Announce authorization required and ask one concise confirmation question before selecting Tier 2 or Tier 3:

    Delegate this review to `oat-reviewer`?
    

  • If the user authorizes delegation and Codex role prerequisites are satisfied, use Tier 1.

  • If the user declines delegation, continue with the existing Tier 2 / Tier 3 fallback flow.

• If the runtime can dispatch reviewer work (subagent_type in Claude Code, Cursor invocation via /name or natural mention, or Codex multi-agent spawn/auto-spawn) → Tier 1.

• If the Task tool is not available or subagent dispatch is not supported → Tier 2.

• If user explicitly requests inline or confirms they are already in a fresh session → Tier 3.

• If REVIEW_INVOCATION=gate and Tier 1 is unavailable, skip Tier 2 and use Tier 3 inline. Do not return fresh-session instructions from a gate invocation.

Step 6b: Tier 1 — Subagent (if available)

First, pre-compute the review artifact path using Step 7 naming conventions so it can be passed to the subagent.

Then spawn the reviewer:

• Use provider-appropriate dispatch:
  • Claude Code: Task tool with subagent_type: "oat-reviewer" (resolves from .claude/agents/oat-reviewer.md).
  • Cursor: explicit invocation /oat-reviewer (or natural mention) with agent resolved from .cursor/agents/oat-reviewer.md or .claude/agents/oat-reviewer.md compatibility path.
  • Codex style: ask Codex to spawn agent(s) for review work and wait for all results; optionally pin agent_type when a specific built-in/custom role is required.
• Pass the Review Scope metadata block from Step 5 as the prompt
• Include the pre-computed artifact path for the subagent to write to
• If a worktree was resolved in Step 1.5: include the worktree path in the prompt so the subagent writes the artifact to the worktree directory, not the current session's working directory
• Run in background if supported (run_in_background: true)

The oat-reviewer agent definition contains the full review process, mode contract, severity categories, artifact template, and critical rules. No additional instructions need to be injected.

After the subagent completes:

• Verify the review artifact was written to the expected path
• Continue with Step 9 (plan update) and Step 9.5 (commit)

Step 6c: Tier 2 — Fresh Session (recommended fallback)

If subagent not available:

• If user is already in a fresh session (confirmed), proceed to Tier 3.
• If Codex reported authorization required and the user approved delegation, do not use Tier 2. Return to Tier 1 and delegate to oat-reviewer.
• If user prefers fresh session: provide instructions and exit.
• If REVIEW_INVOCATION=gate: do not provide fresh-session instructions; run Tier 3 inline instead.

Instructions for fresh session:

To run review in a fresh session:
1. Open a new terminal/session
2. Run the oat-project-review-provide skill with: code {scope}
3. When complete, return to this session
4. Run the oat-project-review-receive skill

Step 6d: Tier 3 — Inline Reset (fallback)

If user insists on inline review in current session:

• Run "reset protocol":
  1. Re-read required artifacts for current workflow mode from scratch
  2. Read all files in FILES_CHANGED
  3. Apply oat-reviewer checklist inline
  4. Write review artifact

Step 7: Determine Review Artifact Path

If INLINE_ONLY=true (user chose inline-only in Step 1.5): skip this step — no artifact path needed.

Review storage contract:

• Write new review artifacts to the active tracked directory: {PROJECT_PATH}/reviews/
• Do not write new artifacts directly into {PROJECT_PATH}/reviews/archived/
• After oat-project-review-receive consumes a review, that skill moves it into reviews/archived/ for local-only historical storage

Naming convention:

Use a seconds-precision UTC timestamp token (YYYY-MM-DDTHHMMSSZ, from date -u +%Y-%m-%dT%H%M%SZ — the -u and the trailing Z are mandatory), not a date-only or local-time stamp, so that same-scope re-reviews and re-gates within one day never collide and oat review latest orders them by recency:

• Phase review: {PROJECT_PATH}/reviews/pNN-review-YYYY-MM-DDTHHMMSSZ.md
• Task review: {PROJECT_PATH}/reviews/pNN-tNN-review-YYYY-MM-DDTHHMMSSZ.md
• Final review: {PROJECT_PATH}/reviews/final-review-YYYY-MM-DDTHHMMSSZ.md
• Range review: {PROJECT_PATH}/reviews/range-review-YYYY-MM-DDTHHMMSSZ.md
• Artifact review: {PROJECT_PATH}/reviews/artifact-{artifact}-review-YYYY-MM-DDTHHMMSSZ.md

Set oat_generated_at in the artifact frontmatter to the matching full timestamp (YYYY-MM-DDTHH:MM:SSZ). Same-second collisions are effectively impossible for sequential runs; if one occurs, append -v2, -v3, etc.

Important: PROJECT_PATH here must be the resolved path from Step 1.5. If a worktree was detected, this path is relative to the worktree root, ensuring the artifact is written on the correct branch.

mkdir -p "$PROJECT_PATH/reviews"

Step 8: Write Review Artifact (if Tier 3)

If running inline (Tier 3), execute the review and write artifact.

Review checklist (from oat-reviewer):

1. Verify scope (don't review out-of-scope changes)
2. If code review: verify alignment to available requirements sources (spec/design for spec-driven mode; discovery/import reference for quick/import)
3. If code review: verify code quality (correctness, tests, security, maintainability)
4. If artifact review: verify completeness/clarity/readiness of the artifact and its alignment with upstream artifacts
5. Categorize findings (Critical/Important/Medium/Minor)
6. For final scope: explicitly disposition deferred Medium ledger items (fix now vs accept defer)
7. Write artifact with file:line references and fix guidance

Review artifact template: (see .agents/agents/oat-reviewer.md for full format)

Shared ad-hoc companion reference (non-project mode):

• .agents/skills/oat-review-provide/references/review-artifact-template.md

---
oat_generated: true
oat_generated_at: { full UTC timestamp, e.g. 2026-07-06T11:16:01Z }
oat_review_scope: { scope }
oat_review_type: { code|artifact }
oat_review_invocation: { manual|auto|gate }
oat_project: { PROJECT_PATH }
---

# {Code|Artifact} Review: {scope}

**Reviewed:** {today}
**Scope:** {scope description}
**Files reviewed:** {N}
**Commits:** {range}

Frontmatter field: oat_review_invocation

• manual (default): Review was manually triggered by the user. oat-project-review-receive uses standard disposition behavior (user prompts for triage, minors auto-deferred for non-final scopes).
• auto: Review was spawned by the auto-review checkpoint trigger in oat-project-implement. oat-project-review-receive uses relaxed disposition: minors are auto-converted to fix tasks (not deferred), no user prompts for disposition decisions.
• gate: Review was spawned by oat gate review for a workflow gate. Gate-originated reviews use normal stateful review-provide behavior: write the review artifact, update the ## Reviews row, and commit review bookkeeping. The gate CLI maps the review artifact findings to exit status, and oat-project-review-receive dispositions gate reviews autonomously (no user prompts) — selecting convert-to-fix-tasks when the gate blocked at its threshold, or non-pausing judgment-sweep disposition (defer / small-fix-now / reject, with sub-threshold findings recorded durably) when the gate passed.

When oat-project-implement spawns this skill for auto-review at checkpoints, it passes context indicating auto invocation. Set oat_review_invocation: auto in the artifact frontmatter.

When oat gate review invokes this skill, it includes gate-originated context instructing the reviewer to write oat_review_invocation: gate. Honor that instruction in the artifact frontmatter.

For all other invocations (user-triggered, fresh session), use manual.

Gate parsing contract:

• Include either the Findings: {N} critical, {N} important, {N} medium, {N} minor summary line or the standard ## Findings section with ### Critical, ### Important, ### Medium, and ### Minor subsections populated with findings or None.
• Do not omit severity headings merely because a severity has zero findings; oat gate review depends on counts or standard Findings sections to determine whether the review blocks.

Summary

{2-3 sentence summary}

Findings: {N} critical, {N} important, {N} medium, {N} minor

Findings

Critical

{findings or "None"}

Important

{findings or "None"}

Medium

{findings or "None"}

Minor

{findings or "None"}

Spec/Design Alignment

Requirements Coverage

Requirement	Status	Notes
{ID}	implemented / missing / partial	{notes}

Extra Work (not in requirements)

{list or "None"}

Verification Commands

{commands to verify fixes}

Recommended Next Step

Run the oat-project-review-receive skill to convert findings into plan tasks.

### Step 9: Update Plan Reviews Section

After review artifact is written, update `plan.md` `## Reviews` table _if plan.md exists_.

Update or add a row matching `{scope}`:

- `Scope`: `{scope}` (examples: `p02`, `final`, `spec`, `design`)
  - Phase-range examples such as `p02-p03` are valid code-review scopes and should be preserved exactly.
- `Type`: `code` or `artifact`
- `Status`: `received` (receive-review will decide `fixes_added` vs `passed`; `passed` now requires no unresolved Critical/Important/Medium and final deferred-medium disposition when applicable)
- `Date`: `{today}`
- `Artifact`: `reviews/{filename}.md`

If plan.md is missing (e.g., spec/design review before planning), skip this update and rely on the review artifact + next-step routing.

### Step 9.5: Commit Review Bookkeeping Atomically (Required)

**If `INLINE_ONLY=true`:** skip this step — no artifact was written to disk.

After writing the review artifact and applying the Step 9 Reviews-table update, create an atomic bookkeeping commit.

**If a worktree was resolved in Step 1.5:** run git commands scoped to the worktree (`git -C "$WORKTREE_PATH" ...`) so the commit lands on the worktree branch, not the current session's branch.

**Commit scope:**

- Always include the active review artifact file: `reviews/{filename}.md`
- Include `plan.md` when Step 9 updated the Reviews table
- Do not write or commit new review artifacts directly into `reviews/archived/`
- Do not include unrelated implementation/code files in this commit

**Commit message:**

- `chore(oat): record {scope} review artifact`

**If the user asks to defer commit:**

- Require explicit user confirmation to proceed without commit
- Warn that uncommitted review bookkeeping can desync workflow routing/restart behavior
- In the summary, clearly state: "bookkeeping not committed (user-approved defer)"

### Step 10: Output Summary

**If subagent used (Tier 1):**

Review requested via subagent.

When the reviewer finishes, run the oat-project-review-receive skill to process findings.

**If fresh session recommended (Tier 2):**

For best review quality, run in a fresh session:

1. Open new terminal/session
2. Run the oat-project-review-provide skill with: code {scope}
3. Return here and run the oat-project-review-receive skill

Or say "inline" to run review in current session (less reliable).

**If inline review completed (Tier 3):**

Review complete for {project-name}.

Scope: {scope} Files reviewed: {N} Findings: {N} critical, {N} important, {N} medium, {N} minor

Review artifact: {path} Bookkeeping commit: {sha or "deferred with user approval"}

Next: Run the oat-project-review-receive skill to convert findings into plan tasks.

## Success Criteria

- Active project resolved
- Review type and scope determined
- Target branch and working directory resolved (worktree detection in Step 1.5)
- Commit range identified
- Files changed list obtained
- Review executed (subagent, fresh session guidance, or inline)
- Review artifact written to the correct branch's working directory (worktree path if applicable; inline-only if user chose that option)
- Plan.md Reviews section updated
- Review artifact + plan bookkeeping committed atomically on the correct branch (or explicitly deferred with user approval)
- For final scope, deferred findings ledger included in reviewer context
- User guided to next step (`oat-project-review-receive`)