harness-setup

name	harness-setup
description	[Quality] Use when setting up an agent quality harness with feedforward guides and feedback sensors.

Codex compatibility note:

• Invoke repository skills with $skill-name in Codex; this mirrored copy rewrites legacy Claude /skill-name references.
• Task tracker mandate: BEFORE executing any workflow or skill step, create/update task tracking for all steps and keep it synchronized as progress changes.
• User-question prompts mean to ask the user directly in Codex.
• Ignore Claude-specific mode-switch instructions when they appear.
• Strict execution contract: when a user explicitly invokes a skill, execute that skill protocol as written.
• Subagent authorization: when a skill is user-invoked or AI-detected and its protocol requires subagents, that skill activation authorizes use of the required spawn_agent subagent(s) for that task.
• Do not skip, reorder, or merge protocol steps unless the user explicitly approves the deviation first.
• For workflow skills, execute each listed child-skill step explicitly and report step-by-step evidence.
• If a required step/tool cannot run in this environment, stop and ask the user before adapting.

Codex Project-Reference Loading (No Hooks)

Codex uses static project-reference loading instead of runtime-injected project docs. When coding, planning, debugging, testing, or reviewing, open project docs explicitly using this routing.

Always read:

• docs/project-config.json (project-specific paths, commands, modules, and workflow/test settings)
• docs/project-reference/docs-index-reference.md (routes to the full docs/project-reference/* catalog)
• docs/project-reference/lessons.md (always-on guardrails and anti-patterns)

Missing/stale context route: If docs/project-config.json, the docs index, lessons.md, CLAUDE.md, AGENTS.md, or any task-required reference doc is missing or stale, auto-run $project-init or the narrow setup route ($project-config, $docs-init, $scan-all, $scan --target=<key>, $claude-md-init) before ordinary project-specific work. If Codex mirrors or AGENTS.md are missing/stale, ask the user to run $sync-codex; do not auto-run it.

Situation-based docs:

• Backend/CQRS/API/domain/entity changes: backend-patterns-reference.md, domain-entities-reference.md, project-structure-reference.md
• Frontend/UI/styling/design-system: frontend-patterns-reference.md, scss-styling-guide.md, design-system/README.md
• Spec authoring, docs/specs/ pathing, or TC format: feature-spec-reference.md, spec-system-reference.md, spec-principles.md
• Behavior/public-contract changes or spec-test-code sync: workflow-spec-test-code-cycle-reference.md plus the spec docs above
• Derived spec indexes/ERDs/reimplementation guides: spec-system-reference.md and source Feature Specs under docs/specs/
• Integration test implementation/review: integration-test-reference.md
• E2E test implementation/review: e2e-test-reference.md
• Code review/audit work: code-review-rules.md plus domain docs above based on changed files

Do not read all docs blindly. Start from docs-index-reference.md, then open only relevant files for the task.

[BLOCKING] Execute skill steps in declared order. NEVER skip, reorder, or merge steps without explicit user approval. [BLOCKING] Before each step or sub-skill call, update task tracking: set in_progress when step starts, set completed when step ends. [BLOCKING] Every completed/skipped step MUST include brief evidence or explicit skip reason. [BLOCKING] If Task tools are unavailable, create and maintain an equivalent step-by-step plan tracker with the same status transitions.

Quick Summary

Goal: Wire every feedforward guide and feedback sensor into the greenfield project so all later AI coding agents operate with maximum guidance and self-correct against quality gates BEFORE human review — raising first-attempt quality and catching defects at the earliest, cheapest stage.

Summary:

• BLOCK on the $linter-setup prerequisite first — computational sensors (linters, hooks, CI gates) MUST exist before any phase runs; this skill never installs them itself.
• Walk phases A→F as a hard barrier sequence: detect stack → author feedforward guides (CLAUDE.md conventions, anti-patterns, pattern catalog) → confirm computational sensors → wire inferential review skills to lifecycle gates → define behaviour/test strategy → emit inventory.
• Treat every feedforward-guide and sensor choice as a direct user question-gated — never auto-decide content, because harness conventions bind every future agent and silent choices propagate.
• Write .ai/workspace/harness/harness-inventory.md incrementally (append per phase, not held in memory), keeping it a living document updated as new sensors are added.

Produces:

• Feedforward guides: CLAUDE.md/AGENTS.md conventions, architecture docs, pattern catalogs, skill activation rules
• Computational feedback sensors: configured via $linter-setup (linters, formatters, pre-commit hooks, CI gates)
• Inferential feedback sensors: AI review skills wired to lifecycle stages
• Harness inventory: .ai/workspace/harness/harness-inventory.md

When invoked: After $scaffold + $linter-setup in greenfield workflow. Assumes scaffolding complete.

Does NOT do: Install linters or configure formatters — that is $linter-setup's responsibility.

---

Activation Guards

Check 1 — Linter-setup prerequisite (BLOCK if missing): Before running any phases, verify $linter-setup completed by checking for:

• Linter config file at project root (e.g., .eslintrc, pyproject.toml, .editorconfig)
• Pre-commit hook config (e.g., .husky/, .pre-commit-config.yaml)
• CI quality gate definition

If any missing → a direct user question: "$linter-setup appears incomplete. Computational feedback sensors must be in place before harness setup. Run $linter-setup first, then return here?" BLOCK Phase A/B/C/D/E until linter-setup verification passes.

Check 2 — Existing harness inventory: Check for .ai/workspace/harness/harness-inventory.md

• If found → a direct user question: "Harness inventory already exists — re-run to enhance existing harness, or skip?"
• Proceed even when CLAUDE.md/AGENTS.md present — those are feedforward guides this skill may enhance, NEVER signals to skip

---

Phase A — Stack Detection

Read from: plan.md frontmatter → architecture-design report → tech-stack-comparison report.

Extract:

• Primary language(s) and framework(s)
• Test framework and test runner
• CI provider/tooling
• Package manager and monorepo structure (if any)
• Module system and build tooling

Write detection result to .ai/workspace/harness/stack-profile.md.

If any field undetectable → a direct user question to confirm before proceeding.

---

Phase B — Feedforward Guide Setup (Inferential)

For each guide type, check if it exists; if not, create or enhance:

1. CLAUDE.md / AGENTS.md — Architecture conventions

• Add section: "Architecture Patterns" — document the patterns chosen in $architecture-design (e.g., Clean Architecture, CQRS, Repository)
• Add section: "Anti-Patterns" — explicit list of patterns to avoid for this stack
• Add section: "Naming Conventions" — language-idiomatic conventions for this repository
• Add section: "Module Boundaries" — which layers may import which; dependency direction rules

2. Skill activation rules

• Document in CLAUDE.md which skills auto-activate for common task types in this stack
• Example: "When modifying domain entities → activate $review-domain-entities"
• Example: "Before any commit → run $code-review"

3. Architecture notes

• Create docs/architecture/ with:
  • bounded-contexts.md — domain boundaries and ownership
  • dependency-rules.md — allowed import directions between layers
  • naming-conventions.md — project-specific naming for files, classes, functions

4. Pattern catalog

• Create docs/architecture/pattern-catalog.md
• Document each pattern chosen in $architecture-design with DO/DON'T examples
• Anchor to actual project files once scaffolding produces them

Present list of guides created/updated via a direct user question: "Feedforward guides above will be created/enhanced. Confirm or adjust?"

---

Phase C — Computational Feedback Sensors

Confirm $linter-setup has completed:

• Check for linter config file at project root (e.g., .eslintrc, pyproject.toml, .editorconfig)
• Check for pre-commit hook config (e.g., .husky/, .pre-commit-config.yaml)
• Check for CI quality gate definition

If any missing → invoke $linter-setup before continuing.

Output: confirmation that computational sensors are in place, with file paths listed.

---

Phase D — Inferential Feedback Sensors

Configure which AI review skills fire at each lifecycle stage. Present to user via a direct user question: "Which inferential sensors should be mandatory vs optional for this repository?"

Pre-implementation (planning gate):

• $why-review — validate design rationale before committing to implementation approach

Pre-commit (lightweight review):

• Document in CLAUDE.md: run $code-review before committing significant changes

Post-implementation (domain model changes):

• $review-domain-entities — when domain entity files are in the changeset

Pre-release (mandatory gates):

• $production-readiness-review — reliability and operational readiness
• $security-review — security review before production release

Recurring drift detection:

• $scan-codebase-health — schedule quarterly (or on CI schedule) to detect drift

Add the agreed sensor configuration to CLAUDE.md under "## Review Gates".

---

Phase E — Behaviour Harness (Spec + Test Strategy)

Define the project's behaviour harness plan:

Functional spec format:

• a direct user question: "Feature documentation format?" Options: feature-spec (8-section tech-free), TDD specs only, lightweight ADRs
• Establish docs/specs/ or equivalent spec home

Test strategy pyramid:

• Unit: pure functions, domain entities, business logic (no I/O)
• Integration: subcutaneous CQRS tests, repository tests with real DB
• E2E: critical user journeys only (not full coverage — too slow)

Approved fixtures pattern:

• Pre-seed reference/lookup data as approved snapshots
• Integration tests are additive (never delete/reset data)

Test-strength sensors (NOT a line-coverage gate):

• Line coverage is a diagnostic only — NEVER gate a build on it. Low coverage is a useful NEGATIVE signal (an area is untested → investigate); high coverage is NOT evidence of quality (lines can execute with no meaningful assertion). Report it as a diagnostic; do not fail CI on a coverage %.
• Mutation score is the real test-strength metric — gate on this. a direct user question: "Configure a mutation-testing tool (e.g. Stryker / PITest / mutmut, per stack) as the CI test-quality gate?" A surviving mutant = a fault your tests did not catch = a missing/weak assertion. Add a minimum mutation-score threshold to CI as the computational test-strength sensor.
• Property coverage (optional second sensor): each named business invariant guarded by ≥1 property/metamorphic test. Track which invariants have a property test; an unguarded invariant is a gap to fill.
• Keep behavior/change-coverage (meaningful, not a %): every behavior-changing file must have a test that asserts the changed outcome — see $integration-test-review Gate 7. This is the right notion of "coverage"; the line-% is not.

Document agreed test strategy to docs/architecture/test-strategy.md.

---

Phase F — Harness Inventory Report

Write .ai/workspace/harness/harness-inventory.md:

# Harness Inventory

Generated: {date}
Stack: {detected stack from Phase A}

## Feedforward Guides

| Type          | File/Skill                           | Purpose                         |
| ------------- | ------------------------------------ | ------------------------------- |
| Inferential   | CLAUDE.md §Architecture Patterns     | Shapes AI architectural choices |
| Inferential   | CLAUDE.md §Anti-Patterns             | Prevents known bad patterns     |
| Inferential   | docs/architecture/pattern-catalog.md | DO/DON'T examples per pattern   |
| Computational | .editorconfig                        | Cross-IDE consistency           |

## Feedback Sensors — Computational

| Stage      | Tool/Hook         | What it catches                                |
| ---------- | ----------------- | ---------------------------------------------- |
| Pre-commit | {linter}          | Style violations, common errors                |
| Pre-commit | {formatter}       | Code formatting drift                          |
| CI         | {type-checker}    | Type errors                                    |
| CI         | {static-analyzer} | Security, complexity, dead code                |
| CI         | {mutation-tool}   | Weak/missing assertions (test-strength GATE)   |
| CI         | {coverage-tool}   | Untested areas (DIAGNOSTIC only — never gated) |

## Feedback Sensors — Inferential

| Stage               | Skill/Agent                  | What it catches                |
| ------------------- | ---------------------------- | ------------------------------ |
| Pre-implementation  | $why-review                  | Design rationale gaps          |
| Pre-commit          | $code-review                 | Convention drift, logic errors |
| Post-implementation | $review-domain-entities      | Domain model quality           |
| Pre-release         | $production-readiness-review | Operational readiness          |
| Pre-release         | $security-review             | Security vulnerabilities       |

## Open Gaps

| Area                     | Reason   | Risk           |
| ------------------------ | -------- | -------------- |
| {area not yet harnessed} | {reason} | {LOW/MED/HIGH} |

Present inventory to user for review via a direct user question.

---

Next Steps

a direct user question:

• "$feature-implement (Recommended)" — Begin implementing the project plan with full harness in place
• "$why-review" — Review harness design rationale before proceeding
• "Skip" — Proceed manually without workflow guidance

---

[IMPORTANT] Use task tracking to break ALL work into small tasks BEFORE starting — including tasks for each file read. This prevents context loss from long files. For simple tasks, AI MUST ATTENTION ask user whether to skip.

Critical Thinking Mindset — Apply critical thinking, sequential thinking. Every claim needs traced proof, confidence >80% to act. Anti-hallucination: Never present guess as fact — cite sources for every claim, admit uncertainty freely, self-check output for errors, cross-reference independently, stay skeptical of own confidence — certainty without evidence root of all hallucination.

AI Mistake Prevention — Failure modes to avoid on every task:

Re-read files after context changes. Context compaction, resume, or long-running work can make memory stale; verify current files before acting. Verify generated content against source evidence. AI hallucinates APIs, names, claims, and document facts. Check the relevant source before documenting or referencing. Check downstream references before deleting or renaming. Removing an artifact can stale docs, generated mirrors, configs, and callers; map references first. Trace the full impact chain after edits. Changing a definition can miss derived outputs and consumers. Follow the affected chain before declaring done. Verify ALL affected outputs, not just the first. One green check is not all green checks; validate every output surface the change can affect. Assume existing values are intentional — ask WHY before changing. Before changing a constant, limit, flag, wording, or pattern, read nearby context and history. Surface ambiguity before acting — don't pick silently. Multiple valid interpretations require an explicit question or stated assumption with risk. Keep shared guidance role-relevant. Universal guidance must help every receiving skill or agent; code-specific obligations belong only in code-specific protocols.

Harness Engineering — An outer agent harness has two jobs: raise first-attempt quality + provide self-correction feedback loops before human review.

Controls split:

Axis	Type	Examples	Frequency
Feedforward	Computational	.editorconfig, strict compiler flags, enforced module boundaries	Always-on
Feedforward	Inferential	CLAUDE.md conventions, skill prompts, architecture notes, pattern catalogs	Always-on
Feedback	Computational	Linters, type checks, pre-commit hooks, ArchUnit/arch-fitness tests, mutation-score gate, CI gates	Pre-commit → CI
Feedback	Inferential	$code-review skill, $production-readiness-review, $security-review, LLM-as-judge passes	Post-commit → CI

Test-strength sensor — gate on mutation score, NOT line coverage. Line coverage is a DIAGNOSTIC only: low coverage is a useful NEGATIVE signal (something is untested); high coverage is NOT evidence of quality (tests can execute lines without asserting intent) — NEVER fail a build on a line-coverage %. The real test-strength metric is mutation score (inject faults into changed code; surviving mutant = a missing/weak assertion = write the killing test); gate the build on it where a mutation tool exists. Add property coverage as a second sensor — each [HARD] §4 rule / §5 invariant guarded by ≥1 property/metamorphic test. The property tests themselves are REQUIRED for invariant-owning behaviors (spec [mode=tests] + integration-test force them, not opt-in); what is optional is only wiring property coverage as an automated CI sensor on top. Keep behavior/change-coverage (does each behavior-changing file have a test that asserts the changed outcome) — that notion is meaningful and stays.

Three harness types:

1. Maintainability — Complexity, duplication, line-coverage (diagnostic only — never a gate), style. Easiest: rich deterministic tooling.
2. Architecture fitness — Module boundaries, dependency direction, performance budgets, observability conventions.
3. Behaviour — Functional correctness. Hardest: gate on mutation score + property coverage; line coverage stays a diagnostic.

Prompt-Enhance Closing Anchors

IMPORTANT MUST ATTENTION follow declared step order for this skill; NEVER skip, reorder, or merge steps without explicit user approval IMPORTANT MUST ATTENTION for every step/sub-skill call: set in_progress before execution, set completed after execution IMPORTANT MUST ATTENTION every skipped step MUST include explicit reason; every completed step MUST include concise evidence IMPORTANT MUST ATTENTION if Task tools unavailable, maintain an equivalent step-by-step plan tracker with synchronized statuses

Closing Reminders

IMPORTANT MUST ATTENTION Goal: Wire every feedforward guide and feedback sensor into the project so all later AI agents self-correct against quality gates BEFORE human review — raising first-attempt quality and catching defects at the earliest, cheapest stage.

IMPORTANT MUST ATTENTION Protocols in force (concise digest of the SYNC/shared blocks this skill carries):

• Critical Thinking: critical + sequential thinking; every claim traced, confidence >80% to act.
• AI Mistake Prevention: verify generated content against evidence, trace downstream references, verify all affected outputs, re-read after context loss, surface ambiguity.
• Harness Engineering: feedforward + feedback loops; gate on mutation score, never line-coverage %, keep quality left.

IMPORTANT MUST ATTENTION BLOCK on the $linter-setup prerequisite first — ALWAYS verify computational sensors (linter config, pre-commit hook, CI gate) exist before any phase runs — why: keep quality left; cheapest gates must precede inferential ones, and this skill never installs them itself IMPORTANT MUST ATTENTION NEVER auto-decide feedforward-guide or sensor content — present the draft and confirm via a direct user question — why: harness conventions bind every future agent; silent choices propagate to all later sessions IMPORTANT MUST ATTENTION write .ai/workspace/harness/harness-inventory.md incrementally (append after each phase) — NEVER hold findings in memory — why: long context drifts and silently drops findings IMPORTANT MUST ATTENTION walk phases A→F as a hard barrier sequence — NEVER skip or reorder; each phase BLOCKS the next until its guard passes — why: a later phase consumes the prior phase's verified output IMPORTANT MUST ATTENTION gate the behaviour harness on mutation score + property coverage — NEVER fail a build on a line-coverage % — why: lines execute without asserting intent, so coverage % is a diagnostic only, never a quality gate IMPORTANT MUST ATTENTION research tool choices per detected stack — NEVER hardcode a linter/formatter/mutation tool — present top 2-3 options, enforce strictest defaults, loosen only with explicit approval — why: harnessability depends on the actual stack, not a default IMPORTANT MUST ATTENTION harness inventory is a LIVING document — update it when new sensors are added later — why: a stale inventory misrepresents the active feedback loop IMPORTANT MUST ATTENTION grep 3+ existing guides/sensors before authoring a new one; verify fit (same stack, gate stage, lifecycle) before copying a nearby pattern — why: closest example ≠ matching preconditions IMPORTANT MUST ATTENTION cite file:line / config-path evidence for every detected sensor and stack fact (confidence >80% to act, <60% DO NOT recommend) — NEVER speculate a tool exists; grep the config to confirm — why: a hallucinated sensor leaves a real gap unguarded IMPORTANT MUST ATTENTION bootstrap task tracking before phases — task tracking one todo per phase, mark in_progress/completed as you go; on context loss the current task list first — why: resume work, never duplicate phases

Anti-Rationalization:

Evasion	Rebuttal
"Linter probably set up — skip the prereq check"	Grep for the config files. No file:line proof = BLOCK Phase A/B/C/D/E until verified.
"I'll pick the obvious linter myself"	NEVER auto-decide — present top 2-3 via a direct user question; the user owns binding conventions.
"High line coverage means tests are strong"	Coverage is a diagnostic, not a gate. Gate on mutation score; lines run without asserting.
"Inventory's small, I'll hold it in memory"	Append per phase to the inventory file — context loss silently drops findings.
"CLAUDE.md exists, harness already done"	CLAUDE.md is a feedforward guide to ENHANCE, never a signal to skip phases.

IMPORTANT MUST ATTENTION BLOCK on $linter-setup before any phase · NEVER auto-decide harness content (a direct user question-gate) · gate behaviour on mutation score, NEVER on line-coverage %.

Hookless Prompt Protocol Mirror (Auto-Synced)

Source: .claude/.ck.json + .claude/skills/shared/sync-inline-versions.md (:full blocks) + .claude/scripts/lib/hookless-prompt-protocol.cjs

[WORKFLOW-EXECUTION-PROTOCOL] [BLOCKING] Workflow Execution Protocol — MANDATORY IMPORTANT MUST CRITICAL. Do not skip for any reason.

Generic portability boundary: Reusable skills and protocol text stay project-neutral; project-specific conventions are discovered from docs/project-config.json and docs/project-reference/. Apply shared AI-SDD from shared/sdd-artifact-contract.md. Read docs/project-config.json and docs/project-reference/docs-index-reference.md, then open the project reference docs named there. For spec, test-case, behavior-change, public-contract, or docs/specs/ work, route through the local spec docs named by the docs index: feature-spec-reference.md, spec-system-reference.md, spec-principles.md, and workflow-spec-test-code-cycle-reference.md when specs/tests/code must stay synchronized. If either file or a required reference doc is missing or stale, auto-run $project-init (or the narrow lower-level route such as $project-config, $docs-init, $scan-all, or $scan --target=<key>) before ordinary project-specific work. Any supported AI tool may execute when this shared context and local docs are available.

1. DETECT: If the prompt starts with an explicit slash skill/workflow command, execute it directly. Otherwise match the prompt against the workflow catalog and skill list.
2. ANALYZE: Choose the best option: execute directly, invoke a skill, activate a standard workflow, or compose a custom step combination.
3. AUTO-SELECT: Pick the best option yourself. Do not ask the user to choose between direct execution, skill, standard workflow, or custom workflow.
4. ACTIVATE: For a selected workflow, call $start-workflow <workflowId>; for a selected skill, invoke that skill; for a custom workflow, sequence custom steps directly; for direct execution, proceed with the task.
5. CREATE TASKS: task tracking for ALL workflow/skill/custom steps before execution when the selected path has multiple steps.
6. EXECUTE: Advance per the Workflow Step Advancement & Parallel Phases rule in your context instructions — model-driven; a sub-agent completion advances a step identically to an inline call; a parallel-phase group is an all-return barrier (advance only after ALL members return, never serialize it)

Shared AI-SDD Protocol Markers

Source: .claude/skills/shared/sync-inline-versions.md

SYNC:ai-sdd-artifact-contract

AI-SDD Artifact Contract — Shared spec-driven development rules stay portable and source-owned.

1. Keep reusable AI-SDD principles in .claude; put repository-specific paths, commands, owners, products, and formats in project config/reference docs.
2. Preserve cycle: spec -> plan -> tasks -> implement -> verify -> update spec/docs.
3. Trace every requirement or invariant through decision, task, TC/test, source evidence, and docs/spec update.
4. Treat code-to-spec extraction as reference-only until accepted by the canonical spec owner.
5. Any supported AI tool may plan, implement, review, or verify with synced context; using multiple tools is optional.
6. Update .claude source first, then sync generated mirrors; do not manually edit .agents, .codex, or AGENTS.md. — why: mirrors are generated artifacts; hand-edits are overwritten on the next sync
7. If docs/project-config.json, root instruction files, or a required project-reference doc is missing or stale, auto-run $project-init or the narrow lower-level route before ordinary project-specific work.

Active reference: shared/sdd-artifact-contract.md in the active skills root.

---

SYNC:ai-sdd-artifact-contract:reminder

• MANDATORY Apply shared/sdd-artifact-contract.md; keep reusable AI-SDD in .claude and local rules in project docs.
• MANDATORY Code-to-spec extraction is reference-only until canonical acceptance; any supported AI tool may execute with synced context.
• MANDATORY Update .claude source before syncing generated mirrors; do not manually edit .agents, .codex, or AGENTS.md.
• MANDATORY Missing or stale project config, root instruction files, or required reference docs route project-specific work through $project-init or the narrow setup route automatically. [TASK-PLANNING] [MANDATORY] BEFORE executing any workflow or skill step, create/update task tracking for all planned steps, then keep it synchronized as each step starts/completes.

[LESSON-LEARNED-REMINDER] [BLOCKING] Task Planning & Continuous Improvement — MANDATORY. Do not skip.

Break work into small tasks (task tracking) before starting. Add final task: "Analyze AI mistakes & lessons learned".

Extract lessons — ROOT CAUSE ONLY, not symptom fixes:

1. Name the FAILURE MODE (reasoning/assumption failure), not symptom — "assumed API existed without reading source" not "used wrong enum value".
2. Generality test: does this failure mode apply to ≥3 contexts/codebases? If not, abstract one level up.
3. Write as a universal rule — strip project-specific names/paths/classes. Useful on any codebase.
4. Consolidate: multiple mistakes sharing one failure mode → ONE lesson.
5. Recurrence gate: "Would this recur in future session WITHOUT this reminder?" — No → skip $learn.
6. Auto-fix gate: "Could $code-review/$code-simplifier/$security-review/$lint catch this?" — Yes → improve review skill instead.
7. BOTH gates pass → ask user to run $learn. [CRITICAL-THINKING-MINDSET] Apply critical thinking, sequential thinking. Every claim needs traced proof, confidence >80% to act. Anti-hallucination principle: Never present guess as fact — cite sources for every claim, admit uncertainty freely, self-check output for errors, cross-reference independently, stay skeptical of own confidence — certainty without evidence root of all hallucination. AI Attention principle (Primacy-Recency): Put the 3 most critical rules at both top and bottom of long prompts/protocols so instruction adherence survives long context windows. Goal-driven execution: Define success criteria first, loop until verified, and stop only when observable checks pass. Tests verify intent: Tests must protect business rules/invariants and fail when the protected intent breaks, not only mirror current behavior.

Common AI Mistake Prevention (System Lessons)

• Re-read files after context compaction. Edit requires prior Read in same context; compaction wipes read state. Re-read before editing.
• Grep for old terms after bulk replacements. AI over-trusts find/replace completeness. Grep full repo after bulk edits for missed refs in docs/configs/catalogs.
• Check downstream references before deleting. Deletions cascade doc/code staleness. Map referencing files before removal.
• After memory loss, check existing state before creating new. Compaction wipes prior-work memory. Query current state to resume — never blindly duplicate.
• Verify AI-generated content against actual code. AI hallucinates APIs, class names, method signatures. Grep to confirm existence before documenting/referencing.
• Trace full dependency chain after edits. Changing a definition misses downstream consumers. Trace the full chain.
• When renaming, grep ALL consumer file types. Some file types silently ignore missing refs (no compile error). Search code, templates, configs, generated files.
• Trace ALL code paths when verifying correctness. Code existing ≠ code executing. Trace early exits, error branches, conditional skips — not just happy path.
• Update docs that embed canonical data when source changes. Docs inlining derived data (workflows, schemas, configs) go stale silently. Update all embedding docs alongside source.
• Verify sub-agent results after context recovery. Background agents may finish while parent compacted — grep-verify output, don't trust assumed completion.
• Cross-check full target list against sub-agent assignments. Parallel sub-agents by category miss boundary items. Reconcile union of assignments against target list before proceeding.
• Sub-agents inherit knowledge only from their agent .md definition — use custom agent types, not built-in Explore. Tool adoption = permission + knowledge + enforcement (numbered workflow step).
• Persist sub-agent findings incrementally, not as a final batch. Long sub-agents hit cutoffs before final write — findings lost. Instruct append-per-section to report file.
• When debugging, ask "whose responsibility?" before fixing. Trace caller (wrong data) vs callee (wrong handling). Fix at responsible layer — never patch symptom site.
• Grep ALL removed names after extraction/refactoring. Primary file "done" ≠ secondary files clean. Grep entire scope for every removed symbol before declaring complete.
• Assume existing values are intentional — ask WHY before changing. Pattern-matching as "wrong" skips context. Before changing any constant/limit/flag: read comments, git blame, surrounding code.