Skip to content

Add outcome-first artifact acceptance gates to Codex and AI-OS workflows #157

Description

@sergstack

Goal

Prevent technical pass reports from being treated as successful delivery when the user-facing business artifact is incomplete or unusable.

Why

A recent Analytics/Codex workflow produced a generated Excel report where technical checks and pipeline status appeared successful, but the business deliverable was not accepted because required proposed fields were empty.

The system needs an explicit outcome-first gate:

Technical checks passed does not equal business acceptance.

Scope

Update existing workflow and handoff documentation only. Keep the change minimal and additive.

Allowed files

  • GOAL_MODE.md
  • .github/ISSUE_TEMPLATE/goal.md
  • .github/ISSUE_TEMPLATE/codex-task.md
  • docs/AI_DEVELOPMENT_WORKFLOW.md
  • docs/templates/CODEX_ISSUE_EXECUTION_PROMPT.md
  • ChatGPT/[AI OS]/Knowledge/HANDOFF_PROTOCOL.md
  • ChatGPT/[AI OS]/Knowledge/GITHUB_ISSUE_DRIVEN_HANDOFF.md
  • Optional, only if needed for consistency:
    • ChatGPT/[AI OS]/PROJECT_INSTRUCTIONS.md

Forbidden changes

  • Do not add automation.
  • Do not add autonomous loops.
  • Do not add semantic search, embeddings, vector DB, web UI, autonomous retrieval, or production agentic workflow.
  • Do not change business logic, formulas, schemas, metrics, or source workbook behavior.
  • Do not create new broad workflow files unless necessary.
  • Do not weaken existing checks.
  • Do not remove no-auto-merge or human-review requirements.
  • Do not add runtime artifacts, logs, raw dumps, secrets, private files, embeddings, or vector DB files.

Required changes

1. Add Outcome-first acceptance rule

Add a concise rule to GOAL_MODE.md:

For any task that produces a user-facing artifact or business deliverable, success requires business acceptance and artifact/content verification, not only passing tests or opening a PR.

Technical checks passed, pipeline completed, files generated, or PR opened are not sufficient if the user-facing result does not satisfy the business outcome.

2. Update Goal issue template

In .github/ISSUE_TEMPLATE/goal.md, strengthen Done when with:

  • Business outcome / Done when
  • Artifact/content checks, if applicable
  • Non-acceptance examples

3. Update Codex strict issue template

In .github/ISSUE_TEMPLATE/codex-task.md, split acceptance into:

  • Business acceptance
  • Artifact/content acceptance
  • Technical checks
  • Non-acceptance examples

4. Update Codex execution prompt

In docs/templates/CODEX_ISSUE_EXECUTION_PROMPT.md, require Codex final reports to include:

  • Business acceptance
  • Artifact/content verification
  • Technical checks
  • Acceptance status
  • No auto-merge

Add a rule:

Do not report success if technical checks pass but the business artifact or user-facing deliverable is not verified.

5. Update AI development workflow

In docs/AI_DEVELOPMENT_WORKFLOW.md, update acceptance section so a PR can be merged only when:

  • scope matches the Issue;
  • allowed files rule is respected;
  • required checks pass;
  • user-facing business acceptance is satisfied, when applicable;
  • artifact/content checks pass, when applicable;
  • risks are documented;
  • human reviewer accepts the change.

6. Update AI-OS handoff docs

In ChatGPT/[AI OS]/Knowledge/HANDOFF_PROTOCOL.md and ChatGPT/[AI OS]/Knowledge/GITHUB_ISSUE_DRIVEN_HANDOFF.md, add outcome-first handoff fields:

  • Business acceptance
  • Artifact/content checks
  • Non-acceptance examples

Acceptance criteria

  • GOAL_MODE.md contains outcome-first acceptance rule.
  • Goal issue template asks for business outcome and non-acceptance examples.
  • Codex task template separates business acceptance, artifact/content acceptance, and technical checks.
  • Codex execution prompt requires final report to distinguish business acceptance from technical checks.
  • AI development workflow states that checks passing alone is not enough for merge/acceptance when a user-facing artifact exists.
  • AI-OS handoff docs include business acceptance and artifact/content checks.
  • Existing no-auto-merge and human-review rules remain intact.
  • No blocked promotion items are introduced.
  • No unrelated cleanup or refactor is performed.

Checks to run

git status --short
python3 scripts/check_project_instructions_length.py
python3 scripts/check_repo_public_safety.py
python3 scripts/check_manifest_paths.py
python3 scripts/check_knowledge_bundles.py
git diff --check

Expected PR summary

The PR must include:

  • issue reference;
  • files changed;
  • where the outcome-first gate was added;
  • checks run;
  • result;
  • residual risks;
  • owner review needed;
  • explicit note: Do not merge automatically.

Risks

  • If the rule is too broad, Codex may over-report artifact checks for tasks that do not produce artifacts.
  • If the rule is too narrow, the contractors Excel failure mode may repeat.
  • Keep wording concise to avoid bloating Project Instructions.

Acceptance status

Ready for Codex implementation as a documentation/governance PR.

Do not merge automatically.

Metadata

Metadata

Assignees

No one assigned

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions