.claude/skills/codex-task-execution-loop/SKILL.md
Single-agent execution loop for implementation tasks. Replaces task-executor/quality-fixer subagent cycle with direct Codex skill-driven implementation and quality gates.
npx skillsauth add ssaattww/excelreport codex-task-execution-loopInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
This skill follows the non-entry execution contract standard.
Required contract_extensions: task_unit_id, cycle_index.
See codex-execution-contract.md and non-entry-execution-contract-template.md for full rules.
input:
objective: "Implement one task unit and run quality gate"
contract_extensions: { task_unit_id: "task-12.3", cycle_index: 2 }
output:
status: "completed"
quality_gate:
gate_id: "task-cycle-check"
gate_type: "implementation"
trigger:
event: "post-cycle validation"
source: "codex-task-execution-loop"
criteria:
- id: "task-unit-complete"
description: "Task unit checks are complete"
- id: "contract-aligned"
description: "Cycle contract fields remain aligned"
result: "pass"
evidence:
- check_id: "cycle-checks"
status: "pass"
summary: "Cycle checks passed"
source_ref: ".claude/skills/codex-task-execution-loop/SKILL.md"
blockers: []
branching:
on_pass: "handoff"
on_fail: "revise"
max_cycles: 2
contract_extensions: { task_unit_id: "task-12.3", cycle_index: 2 }
codex-execution-contract.md.status, summary, changed_files, tests, quality_gate, blockers, next_actions.../workflow-entry/references/non-entry-execution-contract-template.md (objective, scope, constraints, acceptance_criteria, allowed_commands, sandbox_mode) before proceeding.contract_extensions: task_unit_id, cycle_index.non-entry-execution-contract-template.md.coding-principles.testing-principles.Escalate to user if any condition is true:
[Stop: requirement-change-detected] + [Approve: route-selection].[Stop: quality-gate-failed] + [Approve: resume-after-fix] with blocker details.[Stop: high-risk-change] + [Approve: high-risk-change].Run and fix until stable:
Use ai-development-guide for anti-pattern checks and fail-fast error handling discipline.
[Stop: quality-gate-failed] + [Approve: resume-after-fix], then return explicit blockers and next action options.integration-e2e-testing when adding or changing cross-component behavior.quality_gate emission and branching using quality-gate-evidence-template.md.gate_id, gate_type, trigger, criteria, result, evidence, blockers, branching.gate_type: implementation for task execution loops.result: pass|fail|blocked before handoff.result: blocked, emit [Stop: quality-gate-failed] and pause for escalation handling.Use canonical markers: [Stop: <Gate Name>].
Classify every stop as approval_gate or escalation_gate.
At each stop, emit a full gate record: gate_name, gate_type, trigger, ask_method, required_user_action, resume_if, fallback_if_rejected.
Default ask_method is AskUserQuestion.
Resume an approval_gate only with explicit user approved: true; resume an escalation_gate only after user direction or reroute.
Respect batch boundary: no autonomous implementation starts before [Stop: pre-implementation-approval] is approved.
Enforce max_revision_cycles: 2; overflow requires human intervention.
Agent-local task/review outcomes never replace user approvals.
Stop points for this skill:
[Stop: pre-implementation-approval] (approval_gate)[Stop: high-risk-change] (approval_gate)[Stop: requirement-change-detected] (escalation_gate)[Stop: quality-gate-failed] (escalation_gate)[Stop: revision-limit-reached] (escalation_gate)Full protocol and payload schema: ../workflow-entry/references/stop-approval-section-template.md.
databases
Unified deterministic entry for workflow requests. Centralizes routing with stop/approval and sandbox controls.
development
Sends commands to another tmux pane. Use when requests include phrases like "run it in another pane," "send via tmux," or "ask Claude Code to execute."
tools
Use when the user asks to run Codex CLI (codex exec, codex resume) or references OpenAI Codex for code analysis, refactoring, or automated editing
development
End-to-end lifecycle orchestration with a single Codex agent. Replaces subagent workflow coordination while preserving scale-based phases, stop points, and quality gates.