菜单
Synthesize the common stuck points across failed decomposition plans and recursive sub-agent reports. Use when the current batch of decomposition plans has failed.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | identify-key-failures |
| description | Synthesize the common stuck points across failed decomposition plans and recursive sub-agent reports. Use when the current batch of decomposition plans has failed. |
Use this skill to turn many failed attempts into reusable guidance for the next planning round.
Read:
failed_pathscounterexamples and toy_examplesfailed_paths so later planning skills can use it.$propose-subgoal-decomposition-plans.Append to failed_paths:
{
"record_type": "key_failures_summary",
"failed_plan_ids": ["..."],
"plan_failures": [
{
"plan_id": "...",
"stuck_points": ["..."]
}
],
"common_failures": ["..."],
"implications_for_next_plans": ["..."]
}
Also append an events record indicating that a new planning round is needed.
memory_searchmemory_appendbranch_updateIf the reports are too weak to identify meaningful common failures, append an events record with event_type="key_failures_inconclusive" and state what information is still missing.
Screen a decomposition plan by first trying to prove all of its subgoals directly, then identifying the key stuck points if the plan does not fully go through. Use when a decomposition plan is created.
Find relevant math results, constructions, examples, counterexamples, and background references for a statement. Use when you need context for a new problem, supporting references for constructing examples or counterexamples, or external results while proving subgoals.
Validate externally referenced theorems by querying arXiv theorem search first and Codex's built-in web search second. Use when a markdown proof cites statements from external papers.
Verify a markdown proof in the order it is written. Use when the task is to check local correctness, theorem applicability, and reasoning gaps statement by statement through a paper-style proof.
Launch one sub-agent per decomposition plan after direct screening has identified the key stuck points for each plan. Use when all current plans have been screened by direct proving, none fully solves the problem, and parallel recursive work is needed.
Construct candidate counterexamples to test a proposed conjecture, lemma, or intermediate claim by keeping the assumptions true while making the claimed conclusion fail. Use when you are stuck in reasoning and want to see where the assumptions take effect and gain intuition, when a proposed conjecture/claim feels fragile or unproved, or when you want to test whether the assumptions can hold while the claimed conclusion fails.