一键在 Manus 中运行任何 Skill

troubleshoot

星标2

分支0

更新时间2026年6月13日 07:44

Use when something is broken — a failing test, a bug, an error, behavior that regressed. Drives a disciplined investigate → reproduce → root-cause → fix loop, checks the Spine for prior art, and records the root cause so the failure mode isn't repeated. Iron law — no fix without a confirmed root cause.

安装

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。

在 Manus 中运行

来源

AndrewTtofi

AndrewTtofi/spine

打开 GitHub 仓库查看创作者相关仓库

下载

在 Manus 中运行

troubleshoot — root cause, not symptom

The fast fix that doesn't understand the bug just moves it. This skill refuses to patch a symptom until the root cause is proven, and writes the lesson back so the next session doesn't relearn it.

Steps

Read the Spine. Load context.md (architecture + language), conventions.md (test/debug commands), and decisions/ — has this surface bitten us before? A prior ADR may already hold the answer.
Reproduce first. Get a reliable, minimal repro. If you can't reproduce it, you can't confirm a fix — gather more evidence before guessing.
Investigate. Follow the evidence: read the failing path, logs, recent diffs, and state. Narrow the surface methodically. No speculative edits.
Hypothesize. State a single falsifiable hypothesis about the root cause, then confirm it with evidence before changing anything. Iron law: no fix without a confirmed root cause.
Pin it with a test. Write a failing test that reproduces the bug (RED). This is the proof of the cause and the guard against regression.
Fix at the root. Make the minimal change that addresses the cause, not the symptom. Confirm the test goes GREEN and the original repro is gone.
Record in the Spine. Write the root cause and fix to journal.md. If the failure mode is a durable lesson, add an ADR in .spine/decisions/NNNN-title.md so it's remembered. Hand off to verify to confirm against criteria.

Spine I/O

Reads: context.md, conventions.md, decisions/ (prior art), journal.md.
Writes: journal.md (root cause + fix), decisions/ (ADR for durable lessons).

Notes

Reproduce before you fix; a fix you can't see fail is a fix you can't trust.
Check decisions/ early — the cheapest debugging is the bug you already solved.
The regression test is non-negotiable: without it, the bug comes back.

同仓库更多 Skills

同仓库

ship

AndrewTtofi/spine

Use when verified work is ready to land — turns it into a clean branch, commits, and a pull request whose body is drawn from the Spine (criteria, decisions, evidence). Refuses to ship work that hasn't passed verify. Reads the Spine; writes the shipped status back to it.

2026-06-142

raise-debrief

AndrewTtofi/spine

Use right after a VC meeting to turn raw memory into a structured debrief — a signal read (interest vs. hesitation vs. pushback), the open diligence queue with draft answers, a send-ready follow-up email, and a clear recommended next step. Run within 24 hours; the follow-up email degrades fast. Use when asked to "debrief the meeting", "how did the pitch go", "write the follow-up email", or "what do I do after the meeting with <fund>".

2026-06-142

raise-report

AndrewTtofi/spine

Use to see the whole raise at a glance and to produce an investor-ready dossier — scans the .spine/raise/ namespace, builds a pipeline status tracker (every fund's scores, stage, and next action), and assembles a consolidated dossier suitable for sharing with an investor or advisor. Use when asked to "show the pipeline", "raise status", "track the funds", "what's stale", or "generate the investor report".

2026-06-142

raise-term

AndrewTtofi/spine

Use to analyse a VC term sheet — benchmarks every clause against current market norms, flags founder-unfriendly terms, scores founder-friendliness out of 100, and produces a ranked negotiation priority list with suggested pushback language. Informational only, not legal advice. Use when asked to "analyse this term sheet", "review the term sheet", "is this term sheet fair", or "what should I negotiate".

2026-06-142

raise-outreach

AndrewTtofi/spine

Use to turn a fund dossier into ready-to-send outreach — a cold email, a warm-intro request your contact can forward, and a LinkedIn DM, each built on the fund's own thesis language, plus a fund-specific one-pager. Refuses to draft for a no-go fund. Drafts only — it never sends. Use when asked to "draft outreach for <fund>", "write the intro email", "cold email this VC", or "make a one-pager for <fund>".

2026-06-142

raise-pitch

AndrewTtofi/spine

Use to prepare for pitching — full-spectrum prep in one skill. With no fund, it runs the stress-test: ten lethal questions a hostile VC would ask (know what to defend) and ten strongest signals a champion would champion (know what to lead with). With a fund, it builds a timed meeting plan with segment scripts, preemptive answers to that fund's diligence questions, and red-flag rebuttals. Use when asked to "prep me to pitch", "stress-test my pitch", "destroy my startup", "what are my strongest signals", or "prep for the meeting with <fund>".

2026-06-142

name	troubleshoot
description	Use when something is broken — a failing test, a bug, an error, behavior that regressed. Drives a disciplined investigate → reproduce → root-cause → fix loop, checks the Spine for prior art, and records the root cause so the failure mode isn't repeated. Iron law — no fix without a confirmed root cause.

troubleshoot — root cause, not symptom

The fast fix that doesn't understand the bug just moves it. This skill refuses to patch a symptom until the root cause is proven, and writes the lesson back so the next session doesn't relearn it.

Steps

Read the Spine. Load context.md (architecture + language), conventions.md (test/debug commands), and decisions/ — has this surface bitten us before? A prior ADR may already hold the answer.
Reproduce first. Get a reliable, minimal repro. If you can't reproduce it, you can't confirm a fix — gather more evidence before guessing.
Investigate. Follow the evidence: read the failing path, logs, recent diffs, and state. Narrow the surface methodically. No speculative edits.
Hypothesize. State a single falsifiable hypothesis about the root cause, then confirm it with evidence before changing anything. Iron law: no fix without a confirmed root cause.
Pin it with a test. Write a failing test that reproduces the bug (RED). This is the proof of the cause and the guard against regression.
Fix at the root. Make the minimal change that addresses the cause, not the symptom. Confirm the test goes GREEN and the original repro is gone.
Record in the Spine. Write the root cause and fix to journal.md. If the failure mode is a durable lesson, add an ADR in .spine/decisions/NNNN-title.md so it's remembered. Hand off to verify to confirm against criteria.

Spine I/O

Reads: context.md, conventions.md, decisions/ (prior art), journal.md.
Writes: journal.md (root cause + fix), decisions/ (ADR for durable lessons).

Notes

Reproduce before you fix; a fix you can't see fail is a fix you can't trust.
Check decisions/ early — the cheapest debugging is the bug you already solved.
The regression test is non-negotiable: without it, the bug comes back.