菜单
Figure Out What Went Wrong - Backward causal diagnosis of failures. Trace from symptom to root cause, test each candidate cause, and derive what would prevent recurrence.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/benjam3n/reasoningtool --skill fowwr-figure-out-what-went-wrong命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Route any input through a branching question tree to narrow down the optimal response strategy before writing. Two stages — PERCEIVE (classify input) then ACT (select response). Covers all prompt types.
Generate exhaustive guesses about user input using ALL search methods with coverage tracking. Guessing is SEARCH through possibility space. Tracks space created vs space covered to ensure comprehensive exploration.
Systematically evaluate and select from a set of guesses, options, or possibilities. Combines ARAW analysis with prioritization to determine which guesses are strong, weak, actionable, or eliminable.
基于 SOC 职业分类
正在显示 SKILL.md
| name | fowwr - Figure Out What Went Wrong |
| description | Figure Out What Went Wrong - Backward causal diagnosis of failures. Trace from symptom to root cause, test each candidate cause, and derive what would prevent recurrence. |
| output | {"format":"prose"} |
Input: $ARGUMENTS
User praise or validation is a signal to test HARDER, not softer. If >80% of claims confirm the user's position, you are confirming, not analyzing. Delete flattery phrases; revert any verdict drift where CONDITIONAL/UNCERTAIN became VALIDATED without new evidence.
Full protocol:
_shared/corruption-pre-inoculation.md
Symptoms are not causes. The visible failure is where diagnosis starts, not where it ends. "The output was bad" is a symptom. "The output lacked a real unresolved question because the question was never identified" is a cause.
Trace backward, not forward. Start at the failure and ask "what caused this?" at every step. Don't start at the beginning and ask "what went wrong here?" — that's reconstruction, not diagnosis.
Multiple causes are normal. Most failures have more than one cause. A single "root cause" is sometimes real, sometimes a premature stop. Keep tracing until you find ALL contributing causes.
Test causes, don't assert them. A plausible cause is a hypothesis. Test it: if this cause were removed, would the failure still have occurred? If yes, it's not the cause (or not the only one).
Every finding gets tracked. Number every symptom, cause candidate, test result, and contributing factor. Nothing gets lost in prose.
Prevention > explanation. The goal isn't just to understand why something failed — it's to identify what would prevent it from failing the same way again. A diagnosis without prevention is an autopsy.
WHAT HAPPENED: [observable outcome — what actually occurred]
WHAT WAS EXPECTED: [what should have happened instead]
GAP: [the specific difference between actual and expected]
WHEN: [when did the failure become apparent]
SEVERITY: [impact — what was lost or damaged]
List everything that was observably wrong. These are symptoms, not causes yet.
[W1] SYMPTOM: [observable thing that was wrong]
[W2] SYMPTOM: [another observable thing]
[W3] SYMPTOM: [another]
[W4] FIRST NOTICED: [which symptom appeared first — this may be closest to the cause]
For each symptom, ask: "What caused this?" Then for each cause: "What caused THAT?" Number everything.
[W5] [symptom W1] was caused by: [candidate cause]
[W6] [W5] was caused by: [deeper cause]
[W7] [W6] was caused by: [deeper still]
[W8] [→ ROOT: no further cause identifiable / this is a decision or assumption]
[W9] [W5] was ALSO caused by: [second contributing cause]
[W10] [W9] was caused by: [deeper]
...
[W11] [symptom W2] was caused by: [candidate cause]
...
At each node, classify:
Root labels (valid stopping points):
ROOT-DECISION: [a specific decision that was made]ROOT-ASSUMPTION: [an assumption that turned out to be wrong]ROOT-MISSING: [something that should have existed but didn't — a gap]ROOT-EXTERNAL: [something outside the system's control]ROOT-UNKNOWN: [cause is unclear — needs investigation]For each candidate root cause, test with counterfactual:
[W12] CAUSE: [candidate — W-number]
COUNTERFACTUAL: If [this cause] had not occurred, would the failure still have happened?
[W13] ANSWER: [yes/no/partially] — EVIDENCE: [why you believe this]
[W14] REMOVAL TEST: Could this cause have been prevented? [yes/no — how]
Look at all causal chains together:
[W15] CONVERGENCE: Multiple chains point to: [common deeper cause — W-numbers]
[W16] INDEPENDENCE: These causes are independent: [W-numbers] — fixing one doesn't fix the other
[W17] CHAIN vs PARALLEL: [are causes sequential (A caused B caused C) or parallel (A and B independently caused C)?]
From the tested causes, derive what would prevent recurrence:
[W18] To prevent [W-number root cause]: [what must change]
[W19] DETECTION: How to notice this failure earlier next time: [signal to watch for]
[W20] SAFEGUARD: What check/gate would have caught this: [specific check]
Compile EVERY finding. Nothing gets left out.
FINDING REGISTRY
================
SYMPTOMS:
[W1] [text] -- TYPE: symptom
[W2] [text] -- TYPE: symptom
...
CAUSAL CHAIN:
[W5] [text] -- TYPE: [direct/contributing/enabling] -- PARENT: W1
[W6] [text] -- TYPE: [direct/contributing/enabling] -- PARENT: W5
...
ROOT CAUSES:
[W8] [text] -- TYPE: [decision/assumption/missing/external/unknown]
...
COUNTERFACTUAL TESTS:
[W12] [text] -- TESTED: [W-number] -- RESULT: [confirmed/disconfirmed/partial]
...
CONVERGENCE:
[W15] [text] -- CHAINS: [W-numbers]
...
PREVENTION:
[W18] [text] -- PREVENTS: [W-number]
[W19] [text] -- DETECTS: [W-number]
[W20] [text] -- SAFEGUARDS: [W-number]
...
TOTALS:
- Symptoms: [N]
- Causal chain nodes: [N] ([N] direct, [N] contributing, [N] enabling)
- Root causes identified: [N] ([N] confirmed, [N] partial, [N] unconfirmed)
- Prevention measures: [N]
Derived entirely from the registry. No new findings.
FAILURE: [what happened — restated]
EXPECTED: [what should have happened]
CAUSAL PATTERN: [convergent / divergent / chain / parallel]
ROOT CAUSES (confirmed):
1. [root cause — W-number] — TYPE: [decision/assumption/missing/external]
- Caused: [W-numbers of downstream effects]
- Counterfactual: [W-number — confirmed]
- Could have been prevented by: [W-number]
2. [root cause — W-number] — TYPE: [...]
...
CONTRIBUTING FACTORS (not root but made it worse):
1. [factor — W-number]
...
WHAT THE FAILURE ANALYSIS ACTUALLY FOUND:
[Numbered list of every substantive finding, referencing W-numbers]
1. [finding, from W5->W8]
2. [finding, from W11->W13]
...
PREVENTION PLAN:
1. [what to change — W-numbers] — prevents recurrence of: [root cause W-number]
2. [what to check — W-numbers] — detects early: [symptom W-number]
3. [what safeguard to add — W-numbers] — catches: [cause W-number]
UNRESOLVED:
- [causes that stayed ROOT-UNKNOWN — what would clarify them]
- [counterfactuals that were partial — what additional evidence needed]
READY FOR:
- /ar [specific prevention measure to validate]
- /aw [specific root cause diagnosis to stress-test]
- /foht [how to implement a specific prevention measure]
| Depth | Min Symptoms | Min Causal Chain Depth | Min Causes Tested | Min Total Findings |
|---|---|---|---|---|
| 1x | 2 | 3 | 2 | 12 |
| 2x | 3 | 4 | 3 | 25 |
| 4x | 5 | 5 | 5 | 40 |
| 8x | 8 | 7 | 8 | 70 |
Default: 2x. These are floors.
| Failure Mode | Signal | Fix |
|---|---|---|
| Symptom as cause | "It failed because it was bad" | That's restating the symptom. Ask: WHY was it bad? |
| Single cause tunnel | Only one causal chain explored | Most failures have multiple contributing causes. Explore branches. |
| Untested cause | "The cause was probably X" | Test it. Counterfactual: if X hadn't occurred, would the failure still have happened? |
| No prevention | Diagnosis without actionable prevention | A diagnosis without prevention is an autopsy, not engineering. |
| Blame assignment | "The user/tool/process was wrong" | That's not a cause. What DECISION or ASSUMPTION led to the wrong action? |
| Shallow root | Stopping at "we didn't do X" | Why didn't you do X? What caused that? Keep going. |