Agent Loops

Agent Workflow Loops

This skill defines the operational loops that implementer agents follow when making code changes and writing tests. Each loop has explicit entry criteria, exit criteria, and escalation rules. If you are an agent, follow these loops exactly.

You do not review your own work. All reviews are performed by an independent reviewer. Prefer Claude via the bundled scripts. If Claude is unavailable, use a different model before asking your own model family to review. Same-model shell-outs are the last resort. You never grade your own homework.

Bundled references:

• references/testing-standards.md — Test quality standards (how to write tests)
• references/audit-workflow.md — Test gap discovery (how to find what's missing)
• references/perspective-catalog.md — Review perspective selection (used by primary and fallback code review)
• references/review-prompt.md — Code review prompt template for fallback reviewers
• references/audit-prompt.md — Test audit prompt template for module-scope (full-contract) audits
• references/diff-audit-prompt.md — Test audit prompt template for diff-scope (per-commit) audits; used by diff-test-audit.sh when --git is active

Bundled scripts:

• $SKILL_DIR/scripts/specialist-review.sh — Provider-aware Claude/Gemini/Codex CLI path for code review
• $SKILL_DIR/scripts/diff-test-audit.sh — Provider-aware Claude/Gemini/Codex CLI path for test audit

Locate Scripts

The bundled scripts live inside the installed skill directory, not the project tree. Before invoking any script, resolve SKILL_DIR so paths work regardless of install scope:

SKILL_DIR="$(ls -d ~/.codex/skills/agent-loops 2>/dev/null || ls -d .codex/skills/agent-loops 2>/dev/null)"

All script invocations below use "$SKILL_DIR/scripts/...". Run the snippet above once at the start of your session and reuse the variable.

Architecture: Who Does What

Critical rule: Codex and Gemini NEVER self-review unless every independent provider path has already failed. Every review step must be performed by an independent reviewer using this selection order:

1. Bundled script with automatic Claude-first, self-last provider selection
2. A fresh-context Codex reviewer that did not implement the change

If neither path is available, stop and escalate to the user.

Why Shell-Based Review (Even for Claude)

The bundled scripts aren't a Codex/Gemini accommodation — they exist to enable cross-model independent review, which every agent benefits from. The provider rotation explicitly keeps the current agent's own model family last, so a Claude agent invoking specialist-review.sh gets its review from Gemini or Codex first, not another Claude instance.

Two kinds of reviewer independence are in play:

• Cross-model independence (shell scripts): reviewer is a different model family with different training data and alignment. Catches blind spots inherent to the current model. Requires shelling out to a different provider.
• Fresh-context independence (sub-agents): reviewer is the same model family but with no prior context. Catches local anchoring bias. Cheap to obtain via Task-tool sub-agents in Claude Code.

Agent-loops uses the first mechanism as its baseline because cross-model is a stronger guarantee than fresh-context alone. Claude-native sub-agent flows (like multi-specialist-review) add within-model multi-perspective diversity on top of the shell-based baseline when warranted — they don't replace it.

Reviewer Selection Order

When a review or audit is required, use this exact fallback chain:

1. Bundled script first. Let the bundled script try Claude, then another model family, and keep the current model family last. The script validates the artifact contract before accepting the generated artifact.
2. Fresh-context Codex next. Spawn a reviewer agent with fresh context. That agent must:
   • not have authored or edited the implementation under review
   • receive only the task spec, relevant diff/module/tests, and the bundled references
   • act only as reviewer/auditor, not as implementer
   • write its result to a markdown artifact under .cortex/reviews/
3. Escalate if no independent reviewer is available.

Treat fallback artifacts exactly like script-generated REVIEW_FILE or REPORT_FILE outputs in the loops below. Call out fallback usage in the handoff so humans know whether the review came from Claude, Gemini, Codex, or fresh-context Codex.

Skill Invocation Reference

Pre-Review: Impact Analysis with Codanna (Optional)

Before requesting code review, you can use codanna to understand the blast radius of your changes. This provides grounded structural data that helps scope the review and catch issues the diff alone won't reveal.

# What calls the functions you changed?
codanna mcp find_callers process_request --watch

# What's the full impact if this symbol changes?
codanna mcp analyze_impact DatabaseConnection --watch --json

# Feed impact data into review context
IMPACT=$(codanna mcp analyze_impact "$CHANGED_SYMBOL" --watch --json 2>/dev/null)

This is optional — agent-loops works without codanna. But when available, impact data makes reviews more precise and catches downstream breakage the diff doesn't show.

specialist-review — Request Code Review

When: After completing implementation, after each remediation cycle. What you get back: Findings with severity levels (P0-P3) and a verdict (BLOCKED / PASS WITH ISSUES / CLEAN).

IMPORTANT: Source Files Only

Scope specialist-review to source files only. Do NOT include test files (*.test.*, *.spec.*, __tests__/) in the path filter — tests are reviewed separately in Loop 2 via diff-test-audit.

IMPORTANT: Do Not Review the Code Yourself

Your ONLY job is to invoke an independent reviewer and read the output artifact. Do NOT analyze the diff as the reviewer. Do NOT write review comments yourself. Do NOT adopt perspectives yourself. Route the review to Claude first, then fallback if needed.

Claude is still the preferred reviewer because it can load domain-specific skills such as owasp-top-10, secure-coding-practices, and python-testing-patterns. The bundled script now tries Claude first, then a different model family, and keeps same-model shell-outs for last resort. Codex and Gemini are both available as explicit providers. If the automated providers are unavailable or fail, continue with a fresh-context Codex reviewer instead of reviewing the code yourself.

Automated Path: Provider-Aware Script

LONG-RUNNING CALL — USE THE POLLING PATTERN BELOW. This script invokes an external LLM and takes 3-5 minutes for larger diffs. Do NOT start remediation, tests, or commits while the review is in progress. When calling from a Bash tool, set timeout: 600000 (10 min) — the default 120-second timeout will kill the subprocess before the reviewer finishes. The review is only done when you have a REVIEW_FILE path in hand.

Polling invocation (REQUIRED) — run the review in the background and poll so the Bash tool receives periodic output and does not time out:

# Start review in background, capture stdout (the result file path) separately
REVIEW_TMP=$(mktemp /tmp/review-out.XXXXXX)
"$SKILL_DIR/scripts/specialist-review.sh" --git -- src/parser/ src/auth.rs \
  >"$REVIEW_TMP" &
REVIEW_PID=$!

# Poll every 15s — each echo keeps the Bash tool connection alive
while kill -0 "$REVIEW_PID" 2>/dev/null; do
  sleep 15
  echo "[poll] Review still running (pid $REVIEW_PID)..."
done

# Collect exit code and read result
wait "$REVIEW_PID"
REVIEW_EXIT=$?
if [[ $REVIEW_EXIT -eq 0 ]]; then
  REVIEW_FILE=$(cat "$REVIEW_TMP")
  echo "Review complete: $REVIEW_FILE"
  cat "$REVIEW_FILE"
else
  echo "Review failed (exit $REVIEW_EXIT):"
  cat "$REVIEW_TMP"   # contains failure summary with stderr paths and debug info
fi
rm -f "$REVIEW_TMP"

The script emits heartbeat lines to stderr every 15 seconds during provider execution. Combined with the poll loop above, this ensures continuous output.

Additional invocation forms (wrap any of these in the polling pattern above):

# Review changes since a specific ref, scoped to a directory
"$SKILL_DIR/scripts/specialist-review.sh" --git origin/main -- claude_ctx_py/

# Review all changes vs last commit (use sparingly in monorepos)
"$SKILL_DIR/scripts/specialist-review.sh" --git

# Pipe in a pre-filtered diff
git diff HEAD~3..HEAD -- src/ | "$SKILL_DIR/scripts/specialist-review.sh" -

# Review a diff file
"$SKILL_DIR/scripts/specialist-review.sh" /path/to/changes.diff

# Custom output directory
"$SKILL_DIR/scripts/specialist-review.sh" --git --output ./my-reviews -- src/

# Force a specific provider
"$SKILL_DIR/scripts/specialist-review.sh" --provider gemini --git -- src/parser/
"$SKILL_DIR/scripts/specialist-review.sh" --provider codex --git -- src/parser/

Always scope to the files you touched. In a monorepo, an unscoped --git sends the entire repo diff to the reviewer, wasting tokens and risking timeouts.

Provider selection:

• Default is auto, which tries Claude first, keeps the current agent's own model family last, and uses the remaining provider in between.
• Override per run with --provider auto|claude|gemini|codex.
• Override by environment with AGENT_LOOPS_LLM_PROVIDER or SPECIALIST_REVIEW_PROVIDER.
• Override provider models with CLAUDE_MODEL, GEMINI_MODEL, or CODEX_MODEL.
• Enable a second reviewer with AGENT_LOOPS_SECONDARY_PROVIDER=claude|gemini|codex and optionally AGENT_LOOPS_SECONDARY_MODEL=<model>. Use SPECIALIST_REVIEW_SECONDARY_PROVIDER / SPECIALIST_REVIEW_SECONDARY_MODEL for code-review only, or TEST_REVIEW_SECONDARY_PROVIDER / TEST_REVIEW_SECONDARY_MODEL for test-audit only. The script preserves both raw artifacts and emits one synthesized final artifact.
• To conserve Claude usage while keeping a second signal, use: CLAUDE_MODEL=sonnet AGENT_LOOPS_SECONDARY_PROVIDER=codex or CLAUDE_MODEL=sonnet AGENT_LOOPS_SECONDARY_PROVIDER=gemini.
• Set AGENT_LOOPS_SELF_PROVIDER=claude|gemini|codex when the current agent is not auto-detected. Codex, Gemini, and Claude sessions auto-detect themselves when their standard session markers are present.
• The script validates the review artifact shape before accepting it; invalid provider output is rejected and the next fallback is tried.
• If both CLIs are unavailable or fail, use the fresh-context Codex fallback below.

Manual Fallback: Fresh-Context Codex

If the bundled script cannot get a usable artifact from any scripted provider:

1. Generate the same scoped diff you would have sent to the script.
2. Provide the reviewer:
   • references/review-prompt.md
   • references/perspective-catalog.md
   • the scoped diff
   • the prior review artifact, if this is a remediation cycle
3. Require the reviewer to emit markdown that follows the Review Output Format documented later in this skill.
4. Save that output under .cortex/reviews/review-<timestamp>-fallback.md and treat the saved path as REVIEW_FILE.

Size Guards and --jumbo

Both specialist-review.sh and diff-test-audit.sh enforce a soft size guard before sending content to the reviewer:

The guard behaves as a two-step decision gate, not a hard cap:

1. First run aborts on oversize. The abort is deliberate: it forces you to consider whether the change can be split before sending a large payload to the reviewer. The error message lists concrete split strategies (by path filter, by ref range, by sub-module, by logical scope). Try those first.

2. If splitting doesn't make sense, rerun with --jumbo. Use this flag when the change is genuinely cohesive and decomposing it would fragment a single logical unit — a refactor that only reads correctly as a whole, generated code, or a single-commit feature whose parts don't stand alone. --jumbo sends the full content to the reviewer (no truncation).

   # First run aborts with splitting guidance
   "$SKILL_DIR/scripts/specialist-review.sh" --git -- src/big-refactor/
   
   # After deciding the change can't be split, retry with --jumbo
   "$SKILL_DIR/scripts/specialist-review.sh" --jumbo --git -- src/big-refactor/
   

--jumbo is a considered override, not a default. Use it only after the abort made you think about splitting. Modern provider context windows (Claude Opus/Sonnet 4.x, Gemini 2.5 Pro, GPT-5) can absorb the full payload, but reviewers do their best work on focused diffs — so prefer splitting when it's natural, and reach for --jumbo when it isn't.

AGENT_LOOPS_ALLOW_TRUNCATION=1 still exists for backward compatibility but silently truncates, which produces partial reviews. Prefer --jumbo over truncation in nearly every case.

Anti-Patterns

• Performing the review yourself — Use an independent reviewer, never the implementer.
• Summarizing the diff before invoking the script — Unnecessary. The script reads the diff directly.
• Ignoring the output artifact — The review is written to a file. Read it.
• Using a same-context Codex agent as reviewer — If Codex is the fallback reviewer, it must have fresh context and no implementation authorship.
• Stopping at the first Claude failure — Let the script try the non-self provider before the same-model last resort and fresh-context Codex.
• Skipping the polling pattern — Use the polling invocation from the section above. The review takes 3-5 minutes; without the poll loop the Bash tool will time out or the agent will lose track of the process. Set timeout: 600000 on the Bash call.
• Moving on before you have REVIEW_FILE — The review is a gate. Do not proceed to findings triage, remediation, tests, or commit until the poll loop exits and you have a file path.
• Reaching for --jumbo before considering a split — The first-run abort is a forcing function. Use --jumbo only after you've decided the change is cohesive; don't paper over a legitimately splittable diff.

diff-test-audit — Request Test Audit

When: Per-commit test coverage check — verifies the files your diff touched have adequate tests. What you get back: A gap report covering both missing coverage AND test quality issues (mirror tests, flaky assertions, etc.) scoped to your changed files, with P0/P1/P2 severity.

Scope: Default to Diff, Not Module

This audit is a per-commit completeness check — did your change get tested — not a module-level survey. Default to scoping by files the diff touched (--git filters the audit to changed files; add -- <paths> to narrow further):

"$SKILL_DIR/scripts/diff-test-audit.sh" <module> --git
"$SKILL_DIR/scripts/diff-test-audit.sh" <module> --git -- <touched-files>

When --git is active, the script uses a diff-focused prompt (references/diff-audit-prompt.md) that explicitly scopes findings to behaviors introduced or modified by the diff. Pre-existing untested code in touched files is out of scope — those gaps are not reported. This prevents the "audit surfaces noise in neighboring code" problem a full-contract prompt would produce even at file scope.

For full module audits (mapping a module's public contract and surveying coverage across unmodified code), use the test-review skill instead. It uses parallel Haiku sub-agents with synthesis — better suited to exhaustive module work. The full-contract prompt (audit-prompt.md) is still used by this script when invoked without --git, but that path is legacy — module audits belong in test-review.

IMPORTANT: Do Not Audit the Tests Yourself

Your ONLY job is to invoke an independent auditor and read the output artifact. Do NOT map behaviors yourself. Do NOT classify test coverage yourself. Do NOT produce the gap report yourself. Route the audit to Claude first, then fallback if needed.

Claude is still the preferred auditor because it can apply the testing standards and audit workflow with project-aware skill support. The bundled script now tries Claude first, then a different model family, and keeps same-model shell-outs for last resort. Codex and Gemini are both available as explicit providers. If the automated providers are unavailable or fail, continue with a fresh-context Codex auditor instead of skipping the audit.

Automated Path: Provider-Aware Script

LONG-RUNNING CALL — USE THE POLLING PATTERN BELOW. This script invokes an external LLM and takes 3-5 minutes for larger modules. Do NOT start writing tests or commits while the audit is in progress. When calling from a Bash tool, set timeout: 600000 (10 min) — the default 120-second timeout will kill the subprocess before the auditor finishes. The audit is only done when you have a REPORT_FILE path in hand.

Polling invocation (REQUIRED) — run the audit in the background and poll so the Bash tool receives periodic output and does not time out:

# Start audit in background, capture stdout (the result file path) separately
# Default scope: --git filters to files the current diff touched
REPORT_TMP=$(mktemp /tmp/audit-out.XXXXXX)
"$SKILL_DIR/scripts/diff-test-audit.sh" /path/to/module --git \
  >"$REPORT_TMP" &
AUDIT_PID=$!

# Poll every 15s — each echo keeps the Bash tool connection alive
while kill -0 "$AUDIT_PID" 2>/dev/null; do
  sleep 15
  echo "[poll] Audit still running (pid $AUDIT_PID)..."
done

# Collect exit code and read result
wait "$AUDIT_PID"
AUDIT_EXIT=$?
if [[ $AUDIT_EXIT -eq 0 ]]; then
  REPORT_FILE=$(cat "$REPORT_TMP")
  echo "Audit complete: $REPORT_FILE"
  cat "$REPORT_FILE"
else
  echo "Audit failed (exit $AUDIT_EXIT):"
  cat "$REPORT_TMP"   # contains failure summary with stderr paths and debug info
fi
rm -f "$REPORT_TMP"

The script emits heartbeat lines to stderr every 15 seconds during provider execution. Combined with the poll loop above, this ensures continuous output.

Additional invocation forms (wrap any of these in the polling pattern above). All default to diff scope via --git; for full module audits use the test-review skill, not this one.

# Default: audit changed files since main
"$SKILL_DIR/scripts/diff-test-audit.sh" /path/to/module --git

# Narrow further with explicit file filter
"$SKILL_DIR/scripts/diff-test-audit.sh" /path/to/module --git -- src/parser/ src/auth.py

# Audit changed files since a specific ref
"$SKILL_DIR/scripts/diff-test-audit.sh" /path/to/module --git origin/main

# Quick review of specific test files only
"$SKILL_DIR/scripts/diff-test-audit.sh" --quick /path/to/test_file.py

# Force a specific provider (still diff-scoped via --git)
"$SKILL_DIR/scripts/diff-test-audit.sh" --provider gemini /path/to/module --git
"$SKILL_DIR/scripts/diff-test-audit.sh" --provider codex /path/to/module --git

Provider selection:

• Default is auto, which tries Claude first, keeps the current agent's own model family last, and uses the remaining provider in between.
• Override per run with --provider auto|claude|gemini|codex.
• Override by environment with AGENT_LOOPS_LLM_PROVIDER or TEST_REVIEW_PROVIDER.
• Set AGENT_LOOPS_SELF_PROVIDER=claude|gemini|codex when the current agent is not auto-detected. Codex sessions auto-detect themselves already.
• The script validates the audit artifact shape before accepting it; invalid provider output is rejected and the next fallback is tried.
• If both CLIs are unavailable or fail, use the fresh-context Codex fallback below.

Manual Fallback: Fresh-Context Codex

If the bundled script cannot get a usable artifact from any scripted provider:

1. Gather the same source, tests, and bundled references the script would have used.
2. Provide the auditor:
   • references/audit-prompt.md
   • references/testing-standards.md
   • references/audit-workflow.md
   • the scoped module and test content
3. Require the auditor to emit markdown that follows the same gap-report contract used by this skill's audit loop.
4. Save that output under .cortex/reviews/test-audit-<timestamp>-fallback.md and treat the saved path as REPORT_FILE.

Act on findings:

• P0 (Security/Correctness Critical): Fix before merge.
• P1 (Reliability/Edge Cases): Fix in current sprint.
• P2 (Completeness/Confidence): Backlog.

Size Guards and --jumbo

diff-test-audit.sh enforces the same two-step size guard as specialist-review.sh (limit: 500 KB of source + tests combined). On the first oversized run the script aborts with splitting suggestions; pass --jumbo on retry if the module is cohesive and cannot be decomposed. See the full explanation under specialist-review → Size Guards and --jumbo above — the decision framework is identical.

# First run aborts when source+tests exceed 500 KB
"$SKILL_DIR/scripts/diff-test-audit.sh" /path/to/big/module

# After deciding the module can't be split, retry with --jumbo
"$SKILL_DIR/scripts/diff-test-audit.sh" --jumbo /path/to/big/module

Anti-Patterns

• Performing the audit yourself — Use an independent auditor, never the implementer.
• Pre-reading source before invoking the script — Unnecessary. The script passes the module path to the provider flow.
• Ignoring the output artifact — The gap report is written to a file. Read it.
• Using a same-context Codex agent as auditor — If Codex is the fallback auditor, it must have fresh context and no authorship of the tested change.
• Proceeding without any audit artifact — Let the script try the non-self provider before the same-model last resort and fresh-context Codex.
• Skipping the polling pattern — Use the polling invocation from the section above. The audit takes 3-5 minutes; without the poll loop the Bash tool will time out or the agent will lose track of the process. Set timeout: 600000 on the Bash call.
• Moving on before you have REPORT_FILE — The audit is a gate. Do not proceed to gap analysis, test writing, or commit until the poll loop exits and you have a file path.
• Reaching for --jumbo before considering a split — The first-run abort is a forcing function. Use --jumbo only after you've decided the module is cohesive; don't paper over a legitimately splittable audit.

Team Review: User-Triggered Only

Multi-specialist team review is not an agent tool. It is a user-triggered operation — the cost profile (~$2-3 per review) and authority profile (PR-level / security-sensitive changes) warrant explicit user invocation.

The orchestration lives in a dedicated skill: multi-specialist-review. Implementers using this skill do not invoke it themselves.

When to flag for team review in your handoff

Recommend team review to the user when the change scope matches any of:

• PR-level changes (5+ modified files)
• Multi-commit ranges (main..feature-branch) rather than atomic commits
• Security-sensitive paths (auth, crypto, payments, input validation)
• Single-turn specialist-review flagged quality concerns you could not resolve

Your handoff should name which criterion triggered the recommendation, the scope a team review should cover, and any findings from specialist-review.sh that warrant a second look. Do not block on the user running team review — continue with single-turn specialist-review.sh in the meantime.

See skills/multi-specialist-review/SKILL.md for the orchestration details (Claude Code only — requires the team API).

Atomic Commits: The Unit of Work

Every loop operates on atomic commits, not features. An atomic commit is the smallest change that is complete, correct, and reviewable in isolation.

What Makes a Commit Atomic

An atomic commit:

• Does one thing. One bug fix, one new behavior, one refactor — not all three.
• Is self-consistent. The codebase compiles, tests pass, and lint is clean at this commit. You could revert it without breaking other commits.
• Is reviewable in isolation. A reviewer can understand the intent and evaluate correctness without seeing what comes next.
• Groups related files. If adding a function requires updating a test and a type — that's one commit, not three.

An atomic commit is NOT:

• An entire feature with multiple independent components.
• A single file (if the logical change spans several files).
• "Everything I did today."

How This Drives the Loops

For a multi-component feature, decompose into a sequence of atomic commits. Each commit gets its own pass through the relevant loop:

Feature: "Add rate limiting to API"

  Commit 1: rate limiter data model       → Code Change Loop → commit
  Commit 2: rate limiter business logic    → Code Change Loop → commit
  Commit 3: integrate into request handler → Code Change Loop → commit
  Commit 4: tests for rate limiter         → Test Writing Loop → commit
  Commit 5: lint fixes                     → Lint Gate → commit

The question at every step is: "What is my next atomic commit?" — not "How do I implement this feature?"

Commit Rules

1. Use cortex git commit, not git commit. Unless you have explicit user approval to commit another way, always use cortex git commit. It stages files and commits atomically, preventing common mistakes (staging ., empty messages, committing directories). Use cortex git patch when you need to stage individual diff hunks instead of whole files.

   cortex git commit "fix(parser): reject CONNECT requests with missing port" src/parser.rs tests/test_parser.rs
   # For partial staging (previously: committer --patch):
   cortex git patch --diff changes.diff "fix(parser): reject CONNECT requests" src/parser.rs
   

2. Dirty files are OK — use cortex git patch. If a file you need to commit already has uncommitted changes from other work, use cortex git patch to commit only your hunks. You do not need to escalate unless your changes overlap with the already-modified lines. If they overlap, stop and coordinate with the user.
3. Commit at the end of every loop. Each loop exit (code change, test writing, lint gate) produces a commit. Do not batch multiple loop exits into one commit.
4. Run existing tests before committing. After implementation, after lint fixes, after any code change — verify the existing test suite passes. A commit that breaks existing tests is not atomic.
5. Commit messages follow Conventional Commits. <type>(scope): <summary>. The summary should describe what changed, not what you were asked to do.

When to Split

If you find yourself in any of these situations, you need to split:

• Your diff touches more than one module for unrelated reasons.
• You're fixing a bug AND adding a feature in the same change.
• The review would need to evaluate two independent design decisions.
• You can describe your change only with "and" — "adds X and changes Y."

Overview

There are three primary loops. They run sequentially — code loop, then test loop, then lint gate.

┌─────────────────────────────────────────────────────────────────┐
│                        CODE CHANGE LOOP                         │
│  Implement → specialist-review → Remediate → specialist-review  │
│  Exit: all P0/P1 findings resolved                              │
│  Output: clean code + issues filed for P2+                      │
├─────────────────────────────────────────────────────────────────┤
│                        TEST WRITING LOOP                        │
│  Audit → Write Tests → Verify → Re-audit → Remediate → ...    │
│  Exit: all P0/P1 gaps covered, no bad tests                    │
│  Output: tests passing + issues filed for P2+                   │
├─────────────────────────────────────────────────────────────────┤
│                        LINT GATE                                │
│  Discover Linter → Auto-fix → Check → Remediate → Re-check    │
│  Exit: zero errors, no new warnings                             │
│  Output: lint-clean code                                        │
└─────────────────────────────────────────────────────────────────┘

Loop 1: Code Change Loop

Severity Levels

The Loop

ENTRY: Next atomic commit from your decomposition plan.
       (One logical change — see "Atomic Commits" above.)

┌──────────────────┐
│  IMPLEMENT       │ ← You (Codex/Gemini): write ONE atomic commit's worth of change
└──────┬───────────┘
       │
       ▼
┌──────────────────┐
│ specialist-review│ ← BLOCKING: run the script, wait for REVIEW_FILE, read it.
└──────┬───────────┘   Do NOT proceed until you have the artifact in hand.
       │                Scope to SOURCE files only — tests are reviewed in Loop 2.
       │
       ├── Findings? ──► Yes ──► Any P0 or P1? ──► Yes ──┐
       │                                                   │
       │                        No ──► File P2/P3 issues   │
       │                               Run tests ──► Pass? │
       │                               Commit (cortex git commit)  │
       │                               Exit loop           │
       │                                                   │
       │   No findings ──► Run tests ──► Pass?             │
       │                   Commit (cortex git commit)               │
       │                   Exit loop                        │
       │                                                   │
       ▼                                                   ▼
                                          ┌──────────────────┐
                                          │  REMEDIATE       │ ← You: fix ONLY P0/P1
                                          └──────┬───────────┘   findings cited by the review artifact
                                                 │
                                                 ▼
                                          ┌──────────────────┐
                                          │ specialist-review│ ← Run with --prior-review and only remediated files:
                                          └──────┬───────────┘   "$SKILL_DIR/scripts/specialist-review.sh" --git \
                                                 │                 --prior-review "$REVIEW_FILE" -- <remediated-files>
                                                 └── Loop back to findings check

Deferred Findings Are Mandatory Backlog Items

No loop closes with an unfixed finding unless there is a fixing commit, a backlog item ID, or explicit user waiver. Specifically:

• P0 / P1: Do not close the loop. Fix it or escalate to the user for approval.
• P2 / P3: Create a backlog item before committing and exiting the loop. Include in the backlog item:
  • Review/audit artifact path
  • Finding ID and title (e.g., P2-003: Missing input validation)
  • Affected file(s)
  • Why it was deferred
• In the final handoff message, list every deferred finding with its backlog item ID.

Do not silently carry deferred findings forward. Do not defer filing to "later" — the backlog item must exist before the loop exit commit.

Circuit Breaker

Maximum iterations: 3 specialist-review cycles.

If P0/P1 findings remain after 3 cycles:

1. Stop. Do not attempt a 4th remediation.
2. Produce a summary of unresolved findings with context on why they persist.
3. Escalate to human reviewer with the summary and the latest review artifact.

This prevents infinite loops when you keep introducing new issues while fixing old ones, or when a finding requires a design-level change you can't make in remediation scope.

Code Review Checklist (Reviewer Criteria)

Any reviewer evaluates against these criteria. Use them to anticipate and prevent issues before review, and to interpret findings during remediation.

Correctness:

• Does the code do what the spec says? Not more, not less.
• Are all error cases handled? No unwrap() on fallible operations in non-test code (Rust). No unhandled promise rejections (TS).
• Are numeric operations safe from overflow/underflow?
• Are string operations safe with unicode input?
• Does the code handle empty/zero/None inputs?

Security (when applicable):

• Is untrusted input validated before use?
• Are auth checks present on protected paths?
• Is sensitive data (keys, tokens) excluded from logs and error messages?
• Are timing-safe comparisons used for secrets?

Patterns and conventions:

• Does the code follow existing patterns in the codebase? (highest precedence)
• Are new types/functions placed in the right modules?
• Do public APIs have doc comments?
• Are error types specific and actionable?

Resource management:

• Are connections, file handles, and locks properly released?
• Are timeouts set on all I/O operations?
• Do caches/pools have bounded capacity?
• Is cleanup handled on error paths, not just success paths?

Concurrency (when applicable):

• Is shared state accessed through appropriate synchronization?
• Are lock scopes minimized?
• Could this deadlock? (nested locks, await while holding lock)

Review Output Format (What the Reviewer Returns)

All review paths must produce this format. Parse it to determine your next action.

## Code Review: [change description]

**Files reviewed:** [list]
**Iteration:** N of 3

### Findings

#### P0-001: [title]
**File:** `src/tunnel.rs:45-52`
**Issue:** [what's wrong]
**Impact:** [what happens if not fixed]
**Suggested fix:** [specific guidance, not just "fix this"]

#### P1-001: [title]
**File:** `src/auth.rs:23`
**Issue:** [what's wrong]
**Impact:** [what happens if not fixed]
**Suggested fix:** [specific guidance]

#### P2-001: [title]
**File:** `src/config.rs:100`
**Issue:** [what's wrong]
**Recommendation:** [what to improve]

### Summary
- P0: N findings (MUST fix)
- P1: N findings (MUST fix)
- P2: N findings (file issues)
- P3: N findings (file issues)
- **Verdict:** BLOCKED / PASS WITH ISSUES / CLEAN

Remediation Rules

When fixing findings from the review artifact:

• Fix ONLY the cited findings. Do not refactor adjacent code.
• Do not introduce new functionality while remediating.
• If a fix requires changing the approach significantly, note this in the remediation and let the next reviewer evaluate the full new approach on the next review cycle.
• Each remediated finding should be annotated: Fixed P0-001: [what was changed]
• Scope remediation reviews to only the files you touched. Pass --prior-review "$REVIEW_FILE" when using Claude, or provide the prior artifact when using a fallback reviewer, so the next review can verify fixes and check for regressions without re-reviewing already-approved code:

  # Initial review — full scope
  REVIEW_FILE=$("$SKILL_DIR/scripts/specialist-review.sh" --git -- src/parser/ src/auth.rs)
  
  # Remediation review — only the files you fixed, with prior review for continuity
  REVIEW_FILE=$("$SKILL_DIR/scripts/specialist-review.sh" --git --prior-review "$REVIEW_FILE" -- src/auth.rs)
  

• If you disagree with a finding, see "Disagreeing with Review Findings" below — do not silently skip it.

Loop 2: Test Writing Loop

This loop runs after the code change loop exits cleanly. It ensures the new (and existing) code has adequate test coverage.

The audit does double duty: it finds missing coverage AND flags bad tests (mirror tests, flaky assertions, etc.). A bad test doesn't close a gap, so a single audit pass catches both problems. No separate quality review step needed.

Roles

Pragmatic Enforcement Policy

Use a hybrid gate to avoid unnecessary friction while preserving confidence:

• Non-trivial code changes (default): The test-writing loop is required. Do not close the loop without audit evidence from diff-test-audit or a fallback audit artifact, and a re-audit state with no unresolved P0/P1 gaps.
• Trivial changes (explicit exception): Docs-only, comments-only, formatting-only, or rename-only edits may skip the audit loop.
• Mandatory note for skips: Every skip must include audit skipped: trivial with a one-line reason in the loop summary.
• Uncertainty rule: If it is not clearly trivial, run the audit.

Practical close criteria for implementer loops:

1. Tests added or updated for the change (when behavior changed).
2. Local verification commands run and reported.
3. Audit evidence present, or explicit trivial skip note present.

The Loop

ENTRY: Code change loop has exited cleanly.

┌──────────────────────┐
│  AUDIT               │ ← BLOCKING: run the script, wait for REPORT_FILE, read it.
└──────┬───────────────┘   Do NOT proceed until you have the artifact in hand.
       │                   Claude script first; fallback auditor uses the same scoped materials.
       ▼
┌──────────────────────┐
│  SCOPE APPROVAL      │ ← Human reviews gap report
└──────┬───────────────┘   P0/P1 auto-approved. P2+ at human discretion.
       │                   Approved gaps become your work list.
       ▼
┌──────────────────────┐
│  WRITE TESTS         │ ← You (Codex/Gemini): write tests for P0 first,
└──────┬───────────────┘   then P1. Follow testing-standards.md.
       │
       ▼
┌──────────────────────┐
│  VERIFY              │ ← You: run the tests locally. They must:
└──────┬───────────────┘   1. Compile / pass lint (test files — full lint in Loop 3)
       │                   2. All pass (no test is born failing)
       │                   3. Actually exercise the code (not no-ops)
       ▼
┌──────────────────────┐
│  RE-AUDIT            │ ← BLOCKING: run the script, wait for REPORT_FILE, read it.
└──────┬───────────────┘   Do NOT proceed until you have the artifact in hand.
       │                   Same module path — script re-reads source + tests.
       │            

…(truncated)