fix(aios-master): flip default to delegation-first, remove "No restrictions" label#542
fix(aios-master): flip default to delegation-first, remove "No restrictions" label#542vxavierr wants to merge 1 commit intoSynkraAI:mainfrom
Conversation
…tions label Resolves the behavioral contradiction identified in issue SynkraAI#527. Three conflicting instructions told @aios-master to execute directly: 1. agent-authority.md: 'Execute ANY task directly | No restrictions' 2. aios-master.md identity: 'Universal executor... executes any task directly' 3. core_principles[0]: 'Execute any resource directly without persona transformation' The first core_principle is the first behavioral instruction the LLM reads — it overrides any delegation rules that appear later in the file. Changes: - agent-authority.md: replace 'No restrictions' row with precise capability table. Separates what @aios-master executes directly (framework ops, orchestration, meta-ops) from what it delegates by default (tasks with a mapped exclusive agent). Adds explicit delegation table with --force-execute escape hatch. Adds hard BLOCK for git push / gh pr. - aios-master.md: update identity to reflect orchestrator (not executor) role. Replace core_principles[0] with MANDATORY PRE-EXECUTION CHECK as the first behavioral instruction — ensuring delegation is always checked before any execution. Remove create-next-story from commands listing (it belongs exclusively to @sm; its presence, even with a note, can be read as authorization to execute). Preserves --force-execute for workflow-engine mode and framework debugging. Closes SynkraAI#527
|
@vxavierr is attempting to deploy a commit to the Pedro Valério Lopez's projects Team on Vercel. A member of the Team first needs to authorize it. |
WalkthroughThe pull request restructures Changes
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Possibly related PRs
Suggested labels
🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
There was a problem hiding this comment.
Welcome to aios-core! Thanks for your first pull request.
What happens next?
- Automated checks will run on your PR
- A maintainer will review your changes
- Once approved, we'll merge your contribution!
PR Checklist:
- Tests pass (
npm test) - Linting passes (
npm run lint) - Commit messages follow Conventional Commits
Thanks for contributing!
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (2)
.aios-core/development/agents/aios-master.md (2)
399-401:⚠️ Potential issue | 🔴 CriticalCritical contradiction undermines delegation-first model.
Line 399 states "All agents - Can execute any task from any agent directly," which directly contradicts the delegation-first behavior established in the identity (line 103: "Does NOT execute specialized agent tasks") and core principles (line 106: "DELEGATION IS THE DEFAULT"). This statement gives the model permission to bypass the mandatory pre-execution check and execute specialized tasks directly, defeating the entire PR objective.
🔧 Proposed fix to align with delegation-first model
**I orchestrate:** -- **All agents** - Can execute any task from any agent directly +- **Framework operations** - Executes agents, tasks, workflows, IDS, and meta-operations directly +- **Specialized tasks** - Routes to exclusive agents per agent-authority.md (override with --force-execute if needed) - **Framework development** - Creates and modifies agents, tasks, workflows (via `*create {type}`, `*modify {type}`)🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.aios-core/development/agents/aios-master.md around lines 399 - 401, The bullet "All agents - Can execute any task from any agent directly" contradicts the delegation-first model; update that bullet so it explicitly forbids direct execution of specialized agent tasks and requires delegation and the pre-execution check instead. Replace or reword the "All agents" line to state agents must delegate specialized tasks to the appropriate specialized agent and follow the existing identity rule "Does NOT execute specialized agent tasks" and the core principle "DELEGATION IS THE DEFAULT" (retain the "Framework development" line as-is if still accurate). Ensure the new text enforces the mandatory pre-execution check and does not grant blanket execution permissions.
302-308:⚠️ Potential issue | 🟡 MinorAdd create-next-story.md to delegated tasks comment for consistency.
The delegated tasks comment lists tasks that are delegated but omits
create-next-story.md, which is in the delegation table (agent-authority.md line 84) as delegating to@sm. For consistency and clarity, it should be included in this comment.📝 Proposed fix to include missing delegated task
# Delegated tasks (Story 6.1.2.3): + # create-next-story.md → `@sm` # brownfield-create-epic.md → `@pm` # brownfield-create-story.md → `@pm` # facilitate-brainstorming-session.md → `@analyst` # generate-ai-frontend-prompt.md → `@architect` # create-suite.md → `@qa` # learn-patterns.md → merged into analyze-framework.md🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.aios-core/development/agents/aios-master.md around lines 302 - 308, The "Delegated tasks (Story 6.1.2.3)" comment in aios-master.md is missing the delegated task create-next-story.md; update that comment list to include the entry "create-next-story.md → `@sm`" so it matches the delegation table (the create-next-story.md delegation to `@sm`) and ensure formatting/ordering is consistent with the existing entries like brownfield-create-epic.md → `@pm` and facilitate-brainstorming-session.md → `@analyst`.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Outside diff comments:
In @.aios-core/development/agents/aios-master.md:
- Around line 399-401: The bullet "All agents - Can execute any task from any
agent directly" contradicts the delegation-first model; update that bullet so it
explicitly forbids direct execution of specialized agent tasks and requires
delegation and the pre-execution check instead. Replace or reword the "All
agents" line to state agents must delegate specialized tasks to the appropriate
specialized agent and follow the existing identity rule "Does NOT execute
specialized agent tasks" and the core principle "DELEGATION IS THE DEFAULT"
(retain the "Framework development" line as-is if still accurate). Ensure the
new text enforces the mandatory pre-execution check and does not grant blanket
execution permissions.
- Around line 302-308: The "Delegated tasks (Story 6.1.2.3)" comment in
aios-master.md is missing the delegated task create-next-story.md; update that
comment list to include the entry "create-next-story.md → `@sm`" so it matches the
delegation table (the create-next-story.md delegation to `@sm`) and ensure
formatting/ordering is consistent with the existing entries like
brownfield-create-epic.md → `@pm` and facilitate-brainstorming-session.md →
`@analyst`.
|
Fechando para revisão interna antes de resubmeter. |
nikolasdehor
left a comment
nikolasdehor
left a comment
There was a problem hiding this comment.
Review — PR #542: flip default to delegation-first
@vxavierr, excelente implementação. Esse PR resolve cirurgicamente os 3 pontos que identifiquei na issue #527 — e o faz da maneira mais limpa possível para um PR 1 incremental.
O que está correto
1. agent-authority.md — substituição da tabela "No restrictions"
A nova tabela de capabilities é precisa e inequívoca. A separação em 3 categorias (executa diretamente / delega por default / bloqueado) é exatamente o que um LLM precisa para tomar a decisão certa. O --force-execute como escape hatch explícito preserva a capacidade do @aios-master sem deixar margem para interpretação ambígua.
A tabela de delegação por tarefa (9 entradas) elimina qualquer ambiguidade — o LLM não precisa inferir quem é o agente correto, está mapeado explicitamente.
2. aios-master.md — identity + core_principles[0]
Exatamente o fix correto. O core_principles[0] como "MANDATORY PRE-EXECUTION CHECK" garante que a primeira instrução comportamental que o LLM lê é "verifique antes de executar", não "execute diretamente". A identity atualizada para "Master orchestrator who routes..." reforça o frame correto desde a primeira leitura.
3. Remoção do create-next-story dos commands
Concordo 100%. A substituição por um comentário # NOTE: Story creation is @sm's exclusive domain é a abordagem certa — remove a ambiguidade sem perder a documentação.
Ponto de atenção (fora do diff, para o PR 2)
O CodeRabbit identificou que na linha ~399 de aios-master.md, a seção "I orchestrate" ainda contém:
- **All agents** - Can execute any task from any agent directly
Isso contradiz o modelo delegation-first estabelecido neste PR. Entendo que está fora do escopo deste diff (PR 1 minimal), mas é importante que o PR 2 corrija essa linha — caso contrário o LLM continua recebendo uma instrução conflitante na seção de orquestração.
Também o comentário de delegated tasks (linhas ~302-308) deveria incluir create-next-story.md → @sm para consistência com a tabela de agent-authority.md. Menor prioridade, mas bom alinhar no PR 2.
Veredicto
O PR está correto, bem delimitado, e resolve o problema mais crítico da issue #527. A abordagem incremental (PR 1 minimal + PR 2 full protocol) é a certa. LGTM.
APPROVE
DevOps Review — @devops (Gage)Veredicto: APPROVED — Ready to merge quando reaberto
Ponto para PR 2: Linha ~399 "All agents - Can execute any task directly" contradiz delegation-first. ⚡ Gage — Repository Guardian |
3 participants
@nikolasdehor obrigado pela análise — você identificou algo que eu tinha deixado passar: a contradição não era só no
agent-authority.md, eram três camadas sobrepostas se contradizendo.Esse PR implementa o PR 1 da abordagem incremental que você sugeriu — os dois arquivos, as correções mais críticas. O PR 2 com os 4 protocolos completos vem na sequência.
O que foi feito
Ponto 1 —
agent-authority.md: substituição da linhaNo restrictionsA linha
| Execute ANY task directly | No restrictions |foi substituída por uma tabela precisa que separa:@aios-masterexecuta diretamente (ops de framework, orquestração, meta-ops)--force-executecomo escape hatch explícitogit push/gh pr)Também adicionei a tabela de delegação por tarefa, que torna o comportamento esperado inequívoco para o LLM.
Ponto 2 —
aios-master.md:identity+core_principles[0]Você apontou que
core_principles[0]é a primeira instrução comportamental que o LLM lê — então é ela que define o default, independente do que vem depois. Troquei porMANDATORY PRE-EXECUTION CHECKcomo primeira entrada, garantindo que a verificação de delegação acontece antes de qualquer execução.A
identityfoi atualizada para refletir o papel de orquestrador, não de executor universal.Ponto 3 —
create-next-storyremovido dos commandsConcordo com a análise: mesmo com uma nota
[DELEGATE to @sm], a presença do comando no listing pode ser lida como autorização. Removi completamente — substituí por um comentário direcionando para@sm *draft {epic-path}.O que fica para o PR 2
O rename de
handoff-protocol.md→delegation-protocol.md(ouauthority-enforcement.md) vai no PR 2, junto com os 4 protocolos completos — faz mais sentido renomear quando o arquivo for introduzido, não antes.Testado em produção
Essa implementação está rodando no nosso HUB há algumas semanas. Referência: vxavierr/aios-hub@4084ec1
Closes #527