Compose structured XML-tagged prompts for Codex GPT-5.6 delegation tasks.
npx skills add https://github.com/em411/supercodex --skill codex-prompting-em411
Please help me install this Agent Skill. Skill: codex-prompting Source: https://github.com/em411/supercodex/tree/main/skills/codex-prompting Command: npx skills add https://github.com/em411/supercodex --skill codex-prompting-em411
<task>, <structured_output_contract>, <verification_loop>, and <grounding_rules> give prompts stable internal structure. - Ready-made recipes: End-to-end templates for diagnosis, narrow fixes, root-cause review, research, and prompt-patching tasks. - Anti-pattern guidance: Concrete before/after examples showing how to fix vague framing, missing output contracts, and unsupported certainty. - Use Case: When the supercodex orchestrator delegates a bug fix to Codex, it assembles a prompt with <task>, <completeness_contract>, <verification_loop>, and <action_safety> so Codex applies a scoped, verified fix without stopping early or refactoring unrelated code. ## Quick Start Ask the orchestrator to compose a Codex GPT-5.6 prompt for a debugging task using the diagnosis recipe with verification and grounding blocks.High-intent search queries and answers about installing and using this skill.
Structure prompts with XML tags: a `<task>` block defining the job, an output contract specifying the response shape, a follow-through policy, and verification or grounding blocks where correctness matters. Keep one clear task per run and state what done looks like.
For debugging, combine `<task>` with `<completeness_contract>`, `<verification_loop>`, and `<missing_context_gating>`. These blocks stop Codex from halting at the first plausible answer and prevent it from guessing missing repository facts.
Use `task --resume-last` for follow-up instructions on the same Codex thread. Send only the delta instruction rather than restating the whole prompt, unless the direction changed materially.
Codex stops early when the prompt lacks a completeness contract or follow-through policy. Add `<completeness_contract>` requiring full resolution and `<default_follow_through_policy>` telling it to proceed with the most reasonable low-risk interpretation.
No. Tighten the prompt contract first: sharpen the output schema, add verification loops, and ground claims in evidence. Raising reasoning or adding long explanations is a last resort after the prompt structure is fixed.
marcusglee11Delegate bounded implementation tasks to the Codex agent via MCP.
sukbearaiDelegates frozen implementation plans to a GPT-5.6 Luna executor via Codex CLI.
derekhu0002Automate agentic engineering workflows with eval-first governance and cost-aware model routing.
phillip-hirschCompose structured prompts for Codex and GPT-5.4 coding tasks.