OpenParable
Skip to content
← Back to skills

Aiox Full Sdc

Development
aiox-full-sdc

Aiox Full Sdc is an Agent Skill for development published in the SynkraAI/aiox-core repository on GitHub: a SKILL.md instruction file an AI agent can load to do this task.

EXECUTE lean Full Story Development Cycle for one story: plan via CLI, run validate → develop → review (QG loop) → close with Sequence Lock, durable progress under .aiox/sdc/. Use when: full-sdc, full cycle, SDC, run story end-to-end.

View on GitHub ↗
QUICK START

How to use this skill

Bring this guide into your coding agent with a prompt tailored to the tool you use.

  1. Open your project in Codex.
  2. Copy and paste the prompt below.
  3. Review files and risks before approving any action.
Prompt to paste
Inspect the Agent Skill at https://github.com/SynkraAI/aiox-core/blob/HEAD/.grok/skills/aiox-full-sdc/SKILL.md and all supporting files. Report provenance, permissions, scripts, dependencies, and security risks. Propose an exact installation into .agents/skills/aiox-full-sdc, but wait for my approval before writing files or running scripts.

Copying this prompt does not install or run the skill. Review third-party files before use.

full-sdc (lean EXECUTE)

Orchestrator that runs the SDC for one story. Atomic skills + task files do phase work; CLI owns plan/verify/progress.

Not in scope: hub full-sdc (~2k LOC), worktree product registry, hub conductor adapters, product harvest trees, product deploy hosts.

Invocation

/full-sdc {story-path} [yolo|interactive]
# or
aiox sdc plan {story-path} --mode yolo

Default mode: interactive.

Automated/yolo dispatch additionally requires a positive explicit AIOX_MODEL_BUDGET_CEILING_USD; the story path and full child intent are validated before every model call.

Before each direct subagent/model spawn, materialize the exact outgoing prompt and context in isolated files and run this mandatory mechanical gate:

aiox sdc preflight "{story-path}" \
  --task "{phase-task}" \
  --budget-usd "$AIOX_MODEL_BUDGET_CEILING_USD" \
  --intent-file "{exact-child-intent-file}" \
  --context-file "{exact-child-context-file}"

Continue only on exit 0. Exit 5 is a governance rejection and the model or subagent must not be invoked. Pass the same bytes from these files to the child; never rebuild or enrich the prompt after preflight.

CLI (mechanical — always use)

# 1) Init / resume durable state
aiox sdc plan {story-path} --mode {yolo|interactive}

# 2) What to run next
aiox sdc next {story-path}

# 3) After each phase completes
aiox sdc verify {story-path} {phase} --mark

# 4) Inspect
aiox sdc status {story-id}

State file: .aiox/sdc/{story-id}/state.json (runtime, gitignored under .aiox/).

Phases: validate → develop → review → apply_qa_fixes (loop) → close

Phase map

#PhaseSkillAgentTask SOT
1validatevalidate-story-draft@povalidate-next-story.md
2developdevelop-story@dev / executordev-develop-story.md
3reviewreview-story@qa / quality_gateqa-gate.md
3bapply_qa_fixesapply-qa-fixes@devapply-qa-fixes.md
4deploy——skip (no deploy config)
5closeclose-story@popo-close-story.md (administrative only)

Skill SOT: .aiox-core/development/skills/<name>/SKILL.md

EXECUTE loop (orchestrator MUST follow)

0. aiox sdc plan {story} --mode {mode}
1. LOOP:
   a. aiox sdc next {story}  → phase + skill path
   b. IF no next phase → DONE (status completed)
   c. Load skill SKILL.md + its task SOT without executing phase work
   d. Materialize the exact payload for inline or spawned phase execution
   e. Run `aiox sdc preflight` over that exact payload; HALT on failure
   f. Execute inline or spawn the phase only after preflight succeeds
   g. IF yolo: autonomous; IF interactive: report + pause on decisions
   h. aiox sdc verify {story} {phase} --mark
   i. IF phase=review and verdict FAIL:
        IF mode=interactive and no explicit fix approval:
             report FAIL and pause (do not invoke apply-qa-fixes)
        IF mode=yolo or explicit fix approval:
             CLI selects apply_qa_fixes automatically
             run apply-qa-fixes skill → verify apply_qa_fixes --mark
             IF apply_qa_fixes verify FAIL:
                  HALT and escalate human without returning to review or incrementing qgIterations
             IF apply_qa_fixes verify PASS:
                  CLI returns to review and increments qgIterations
                  IF the third re-review fails → HALT and escalate human
   j. ELSE IF verify FAIL → HALT (do not advance)
   k. ELSE continue loop
2. QA sets Done on PASS/CONCERNS/WAIVED; close-story only finalizes bookkeeping
3. Push/PR: hand off @devops — never push from this skill

Grok / multi-agent dispatch (preferred when available)

For each phase, spawn the matching persona (or run inline if spawn unavailable):

Phasespawn_subagent type (Grok)Prompt seed
validateaiox-poExecute skill validate-story-draft on {path}; mode={mode}
developaiox-devExecute skill develop-story on {path}; mode={mode}
reviewaiox-qaExecute skill review-story on {path}
apply_qa_fixesaiox-devExecute skill apply-qa-fixes on {path}
closeaiox-poExecute skill close-story on {path}

Before dispatch, declare the shared budget ceiling, bind {path}, materialize the exact child prompt/context, and run the aiox sdc preflight command above against those exact files. Pass those same bytes to the child; never rebuild or enrich the payload after preflight. After each subagent returns: run aiox sdc verify … --mark in the main session (orchestrator owns the lock).

Sequence Lock (soft + CLI)

  1. Phases in order. No N+1 until verify of N passes (or skip).
  2. Only QA review sets Status Done.
  3. Close requires the approved QA verdict and an already-Done story.
  4. QG loop max 3 (maxQgIterations in state).
  5. Anti-self-validation: executor ≠ quality_gate.

Post-phase verification (CLI-enforced)

PhaseOn disk
validateStatus Ready+
developwork evidence; not Done
reviewQA verdict or gate file; approved → Done, FAIL → InProgress
apply_qa_fixesnot Done
closeStatus already Done; administrative finalization only

Modes

ModeBehavior
interactiveReport between phases; stop on blockers
yoloAutonomous between phases; stop only on absolute blockers / circuit breaker

Explicitly non-ported

  • Worktree auto-spawn / registry / GC
  • Full auto-ACK matrix (v1 = CLI state + checklists)
  • Product deploy targets

Strip checklist

  • Skills invoke tasks only
  • CLI First progress/verify
  • No product harvest trees in protocol