Investigate bugs through a resumable evidence-driven debug session workflow.
npx skills add https://github.com/chenziyang110/launchdeck --skill sp-debug-chenziyang110
Please help me install this Agent Skill. Skill: sp-debug Source: https://github.com/chenziyang110/launchdeck/tree/main/.claude/skills/sp-debug Command: npx skills add https://github.com/chenziyang110/launchdeck --skill sp-debug-chenziyang110
.planning/debug/[slug].md so work survives interruptions, with explicit resume, derived-issue, and archive lifecycle handling. - Evidence-First Root Cause Analysis: Requires a confirmed understanding checkpoint, causal map, investigation contract, and falsifiable hypothesis before any code change, rejecting surface-only fixes. - Leader/Subagent Orchestration: Routes investigations as leader-inline, one-subagent, or parallel-subagents based on complexity, with subagents restricted to fact-only evidence collection. - Use Case: A regression appears in a spec-kit project after a refactor. Invoke the skill, confirm the debug checkpoint, let it gather reproduction evidence and logs, confirm the root cause, apply a minimal verified fix, and close the session only after human verification. ## Quick Start Ask the agent to run /sp-debug with a description of the bug you want investigated, or leave it blank to resume the most recent debug session.High-intent search queries and answers about installing and using this skill.
Invoke /sp-debug with a description of the bug, or leave it blank to resume the most recent session. The skill creates or resumes a session file in .planning/debug/ and presents a Debug Understanding Checkpoint before any investigation begins.
It enforces a hard gate: no reproduction, source reads, or code edits until the session records a confirmed understanding, causal map, investigation contract, log plan, and observer framing. Fixes are accepted only after the root cause is confirmed and verified through reproduction and tests.
Yes, it requires a spec-kit project structure with a .specify/ directory. Debug session state is stored under .planning/debug/, and the workflow integrates with the project-cognition compass runtime for advisory navigation.
Yes, the session file in .planning/debug/ is the durable source of truth. On invocation without a new issue, it resumes the active session, restoring the current hypothesis, evidence, eliminated theories, and next action.
Subagents are used when there are two or more independent evidence lanes or broad surface area. They only gather facts such as logs, test results, and code traces; they cannot mutate the session file, declare root cause, or close the session.
The workflow requires a formal human verification stage after agent-side verification. The session closes only after explicit human confirmation or an evidence-backed classification as same, derived, or unrelated issue.
abonetoConvert tasks into dependency-ordered GitHub issues from design artifacts.
yairbedermanStore project-session learnings as reusable lessons with metadata and evidence.
openmeterioLocate and maintain package-level domain documentation in the OpenMeter repository.
FerroxLabsExecute small ad hoc tasks with atomic commits and state tracking.