chore: add long-running AI agent playbook scaffolding

.github/copilot-instructions.md

@@ -200,3 +200,17 @@ npm run lint && npm test -- --coverage && npx playwright test
 - **Sem estimativa de tempo.**
 - **Pergunta apenas em ambiguidade real.**
 - **Paralelismo** — research + read + review independentes rodam simultâneos em Agent Mode.
+
+<!-- codex-long-running-agent-overlay:start -->
+## Universal Long-Running Agent Overlay
+
+This section complements the repository-specific guidance already in this file. If anything here conflicts with the repo-specific rules above, the repo-specific rules win.
+
+- `PRD.md` is the task source of truth for long-running sessions.
+- `PROGRESS.md` is the persistent checkpoint log.
+- `GOAL_RESULT.md` is the final execution report.
+- Before coding, read this file, `PRD.md`, `PROGRESS.md` when it exists, `README.md`, project manifests, tests, and the relevant source folders.
+- Work in small checkpoints, run the smallest relevant validation after each meaningful change, update `PROGRESS.md`, and continue until complete or genuinely blocked.
+- Stop only when the requested work is complete, validation is documented, and `GOAL_RESULT.md` reflects the outcome.
+- Do not rewrite unrelated architecture, fake successful validation, expose secrets, or push without explicit operator instruction for the active session.
+<!-- codex-long-running-agent-overlay:end -->

AGENTS.md

@@ -218,3 +218,17 @@ npm run lint && npm test -- --coverage && npx playwright test
 - **Pergunta apenas em ambiguidade real** do pedido. Não pergunta pra confirmar trabalho de execução.
 - **Paralelo é o padrão** — research + read + review independentes rodam simultâneos.
 - **Hooks do `.claude/hooks/`** rodam automaticamente: post-edit faz lint/format, pre-commit bloqueia commit vermelho.
+
+<!-- codex-long-running-agent-overlay:start -->
+## Universal Long-Running Agent Overlay
+
+This section complements the repository-specific guidance already in this file. If anything here conflicts with the repo-specific rules above, the repo-specific rules win.
+
+- `PRD.md` is the task source of truth for long-running sessions.
+- `PROGRESS.md` is the persistent checkpoint log.
+- `GOAL_RESULT.md` is the final execution report.
+- Before coding, read this file, `PRD.md`, `PROGRESS.md` when it exists, `README.md`, project manifests, tests, and the relevant source folders.
+- Work in small checkpoints, run the smallest relevant validation after each meaningful change, update `PROGRESS.md`, and continue until complete or genuinely blocked.
+- Stop only when the requested work is complete, validation is documented, and `GOAL_RESULT.md` reflects the outcome.
+- Do not rewrite unrelated architecture, fake successful validation, expose secrets, or push without explicit operator instruction for the active session.
+<!-- codex-long-running-agent-overlay:end -->

CLAUDE.md

@@ -224,3 +224,17 @@ npm run lint && npm test -- --coverage && npx playwright test
 - **Pergunta apenas em ambiguidade real** do pedido. Não pergunta pra confirmar trabalho de execução.
 - **Paralelo é o padrão** — research + read + review independentes rodam simultâneos.
 - **Hooks do `.claude/hooks/`** rodam automaticamente: post-edit faz lint/format, pre-commit bloqueia commit vermelho.
+
+<!-- codex-long-running-agent-overlay:start -->
+## Universal Long-Running Agent Overlay
+
+This section complements the repository-specific guidance already in this file. If anything here conflicts with the repo-specific rules above, the repo-specific rules win.
+
+- `PRD.md` is the task source of truth for long-running sessions.
+- `PROGRESS.md` is the persistent checkpoint log.
+- `GOAL_RESULT.md` is the final execution report.
+- Before coding, read this file, `PRD.md`, `PROGRESS.md` when it exists, `README.md`, project manifests, tests, and the relevant source folders.
+- Work in small checkpoints, run the smallest relevant validation after each meaningful change, update `PROGRESS.md`, and continue until complete or genuinely blocked.
+- Stop only when the requested work is complete, validation is documented, and `GOAL_RESULT.md` reflects the outcome.
+- Do not rewrite unrelated architecture, fake successful validation, expose secrets, or push without explicit operator instruction for the active session.
+<!-- codex-long-running-agent-overlay:end -->

GOAL_RESULT.md

@@ -0,0 +1,46 @@
+# Goal Result
+
+## Summary
+
+Describe what was completed.
+
+## Changed Files
+
+- file 1
+- file 2
+
+## Validation Commands
+
+```bash
+command here
+```
+
+## Validation Results
+
+- build: pass/fail
+- tests: pass/fail
+- lint: pass/fail
+
+## Remaining Risks
+
+- risk 1
+- risk 2
+
+## Suggested PR Title
+
+`feat: implement ...`
+
+## Suggested PR Body
+
+```md
+## Summary
+- ...
+
+## Validation
+- [x] tests
+- [x] build
+- [x] lint
+
+## Risks
+- ...
+```

PRD.md

@@ -0,0 +1,60 @@
+# PRD - Feature Name
+
+## Objective
+
+Describe exactly what must be implemented.
+
+## Context
+
+Explain the repository, current behavior, and desired behavior.
+
+## Requirements
+
+- [ ] Requirement 1
+- [ ] Requirement 2
+- [ ] Requirement 3
+
+## Non-Goals
+
+The agent must not:
+
+- rewrite unrelated modules
+- change unrelated APIs
+- introduce large dependencies without need
+- remove tests
+
+## Technical Notes
+
+Relevant files/folders:
+
+```text
+src/
+tests/
+README.md
+package.json
+```
+
+## Validation Commands
+
+Run the available commands:
+
+```bash
+npm test
+npm run build
+npm run lint
+```
+
+Or for .NET:
+
+```bash
+dotnet test
+dotnet build
+```
+
+## Done When
+
+- [ ] All requirements are implemented
+- [ ] Tests pass
+- [ ] Build passes
+- [ ] Lint/typecheck passes, when available
+- [ ] GOAL_RESULT.md is written

PROGRESS.md

@@ -0,0 +1,29 @@
+# Progress Log
+
+## Current Status
+
+Not started.
+
+## Checkpoints
+
+### Checkpoint 1
+
+Status: pending
+
+Task:
+
+Result:
+
+Validation:
+
+Next:
+
+## Blockers
+
+None.
+
+## Validation History
+
+| Command | Result | Notes |
+|---|---|---|
+| | | |