Skip to content
Back to skills

Sw Debug

ASecurity

Investigation-first debugging workflow. Scopes the problem, delegates root cause analysis, produces a diagnosis report, and applies fix/log/defer autonomously.

  • 9 stars
  • 0 votes
  • 0 copies
  • 4 views
  • Added May 26, 2026
developmentgobashdebugging

Security analysis

A100/100

Scanned May 27, 2026

npx -y skills add Obsidian-Owl/specwright --skill sw-debug --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Sw Debug?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Sw Debug
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/obsidian-owl-sw-debug/badge)](https://www.skillsdirectory.com/skills/obsidian-owl-sw-debug)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: sw-debug
description: >-
  Investigation-first debugging workflow. Scopes the problem, delegates root
  cause analysis, produces a diagnosis report, and applies fix/log/defer autonomously.
argument-hint: "[problem description]"
allowed-tools:
  - Read
  - Write
  - Bash
  - Glob
  - Grep
  - Task
---

# Specwright Debug

## Goal

First-class debugging path. Scope the problem, investigate concurrently,
diagnose with evidence, then decide: fix it now, log it, or defer. Operates
autonomously, applying `protocols/decision.md` for the fix/log/defer decision.

## Inputs

- Problem description (argument or recent error context)
- Initial evidence: error messages, logs, failing test output
- `{projectArtifactsRoot}/config.json` — `backlog.type` and `backlog.label`
- Codebase files — read during investigation

## Outputs

- `diagnosis.md` at `{workArtifactsRoot}/{id}/diagnosis.md` — always produced
- `spec.md` at `{workArtifactsRoot}/{id}/spec.md` — Fix path only (2-3 acceptance criteria)
- `decisions.md` at `{workArtifactsRoot}/{id}/decisions.md` — fix/log/defer decision recorded per `protocols/decision.md`

## Constraints

**Stage boundary (LOW freedom):**
Follow `protocols/stage-boundary.md`. Investigate and diagnose. NEVER write code,
run tests, branch, or commit. Fix path: produce spec.md, handoff to `/sw-build`.

**Scope (MEDIUM freedom):**
- If argument provided, use it. If no argument, infer from recent error context
  (last failed command, error output in conversation). If genuinely undetermined,
  apply DISAMBIGUATION: choose the most likely problem from available context.
- Collect evidence: error messages, stack traces, failing tests. Define boundary.

**Investigate (HIGH freedom):**
Delegate concurrently per `protocols/delegation.md`: `specwright-researcher` (code
context, call paths) and `specwright-architect` (root cause, blast radius).

**Diagnose (MEDIUM freedom):**
Write `diagnosis.md`: Problem (observed vs expected), Root Cause (confidence level,
file:line evidence), Blast Radius (affected / not affected), Fix Approach (high-level),
Alternatives Considered. If agents return insufficient evidence: produce low-confidence
diagnosis and note the gap in decisions.md.

**Decision (MEDIUM freedom):**
Apply `protocols/decision.md` DISAMBIGUATION + reversibility:
- Fix spans ≤3 files with local architectural scope → **Fix it now** (Type 2). Write
  spec.md, handoff to `/sw-build`.
- Fix spans >3 files or crosses architectural boundaries → **halt and recommend
  `/sw-design`** (Type 1 — structural scope).
- Known pattern (matches patterns.md entry) → **Log it** as BL-{n} per `protocols/backlog.md`.
- Requires design-level decisions → **Defer** as BL-{n} with `defer` tag.
Record the decision in decisions.md.

**State (LOW freedom):**
Follow `protocols/state.md`. Work ID: `debug-{short-description}`.

## Protocol References

- `protocols/stage-boundary.md` -- scope and handoff
- `protocols/decision.md` -- autonomous decision framework (fix/log/defer)
- `protocols/state.md` -- workflow state updates
- `protocols/delegation.md` -- concurrent delegation
- `protocols/backlog.md` -- backlog items

## Failure Modes

| Condition | Action |
|-----------|--------|
| No problem description or context | Apply DISAMBIGUATION from conversation history. Record interpretation. |
| Agents return no evidence | Low-confidence diagnosis. Apply fix/log/defer per reversibility. |
| Fix is architectural (>3 files) | Halt. Recommend `/sw-design`. |
| Compaction during investigation | Re-run investigation; diagnosis.md rebuilt from scratch |

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.