agentsclimarketplace
Sign in

Debug mode

Skill wangnan0916/cursor-like-debug-mode/skills/debug-mode

Diagnose one reproducible runtime bug with temporary probes, NDJSON evidence, manual checkpoints, and mechanical cleanup.From its SKILL.md

Install
npx -y skills add wangnan0916/cursor-like-debug-mode --skill debug-mode

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

5.6 KB, ~1.1k tokens by cl100k_base, as published. Nobody here has run it

Cursor-like Debug Mode

Run this code-first evidence loop:

analyze code -> hypothesize -> arm -> instrument -> reproduce -> prove -> fix -> verify -> clean

Work autonomously through instrumentation. The first planned user interaction is the Step 5 reproduction checkpoint. Reproduction and verification are gates. Cross them only through the checkpoint protocol.

1. Analyze the Code Path

Start repository legwork from the reported symptom. Read relevant entry points, event handlers, state transitions, existing logs, and recent changes. Map candidate code paths from the user's trigger to the observed outcome. Write a provisional Bug Card from available evidence:

  • Symptom and expected behavior
  • Known trigger, route, command, input, or fixture
  • Candidate runtime path and disputed boundaries
  • Relevant environment, frequency, recent changes, and evidence
  • Observable acceptance criterion

Record unavailable facts as unknown. Resolve consequential unknowns through code reading or discriminating runtime probes.

Complete when: one candidate path links trigger to symptom, and every disputed branch has a named probe boundary.

2. Build the Hypothesis Matrix

Write two to five ranked hypotheses before adding probes. Anchor the first matrix in code analysis and available evidence. Use this shape:

H1 <short cause>
Prediction: <runtime state, path, or timing>
Probe: <event names and code locations>
CONFIRMED when: <specific record condition>
REJECTED when: <specific record condition>

A hypothesis without a concrete prediction requires more code reading. Prefer probes that separate multiple hypotheses.

Complete when: every hypothesis has falsifiable confirmation and rejection conditions. Every hypothesis also has a distinguishing planned probe.

3. Arm the Collector

Read references/collector.md before running any collector command. Start or reuse the collector through its safe wrapper. Capture all five printed collector values.

Complete when: DEBUG_URL, SESSION_ID, LOG_DIR, LOG_FILE, and HEALTH_URL are known.

4. Instrument the Boundaries

Read references/probes.md before editing instrumentation. Add three to eight probes at boundaries that distinguish the live hypotheses. Use runId: "pre-fix" and the collector's session. Wrap every helper and probe in the exact region markers. Inspect every payload field against the privacy allowlist.

Send events to DEBUG_URL + "/log". Runtime standard output stays unchanged.

Complete when: every live hypothesis has a distinguishing probe. Every probe is marked, session-correct, low-volume, and privacy-reviewed.

5. Cross the Reproduction Gate

Read references/checkpoints.md before presenting the pre-fix checkpoint. Execute the pre-fix protocol exactly. Then wait and route the result through its state table.

Complete when: a valid A, B, or C result has been routed by phase.

6. Prove the Cause

After pre-fix A, summarize the log before editing product code:

node <skill-dir>/scripts/summarize-log.mjs <LOG_FILE> --expect-probe <probe_name>

Repeat --expect-probe for every expected event. Inspect raw NDJSON wherever the summary loses ordering or detail. Classify every hypothesis as CONFIRMED, REJECTED, or INCONCLUSIVE. Write the evidence report in this shape:

H1 <short cause>
Status: CONFIRMED
Evidence: <specific events and values>

H2 <short cause>
Status: REJECTED
Evidence: <specific events and values>

Acceptance: <one observable sentence>
Next action: <smallest correction for the confirmed cause>

Missing expected probes block confirmation. If every result is rejected or inconclusive, return to Step 2 with narrower hypotheses.

Complete when: current-run records confirm a cause that explains the symptom. The acceptance criterion is stated in one sentence.

7. Make the Evidence-Bound Fix

Make the smallest change that addresses the confirmed cause. Keep every probe active. Change probe events to runId: "post-fix". Add a focused regression test when the behavior is automatable.

Complete when: each product edit traces to the confirmed cause. Post-fix events and the acceptance check are ready.

8. Cross the Verification Gate

Present the post-fix checkpoint from references/checkpoints.md. Then wait and route the result through its state table.

Complete when: the user returns post-fix B for the stated acceptance criterion.

9. Clean Mechanically

Remove every paired region marked DEBUG_MODE_PROBE <session_id>. Search for the session ID, DEBUG_MODE_PROBE, and both region marker forms. Remove copied helpers and other temporary instrumentation. Delete LOG_FILE by default. Retain it when the user requests local evidence. Keep the shared collector running. Stop it upon explicit user request. Run the relevant checks and inspect the final diff.

Complete when: marker and session searches return zero repository matches. Checks pass, evidence is handled, and only product changes remain.

Exception Branch

Read references/troubleshooting.md only when collection, reproduction, or browser delivery fails.

What ships with it: 10 files

33.1 KB alongside SKILL.md, 5 of them executable

agents/

assets/

references/

scripts/

Gives 0 of the 12 instructions most debug triage skills give in ~1.1k tokens

Counted across 839 of the 1,149 authors here whose files we hold, read 2026-08-07

  • Investigate root cause before proposing any fixin 102 of 839, across 67 files
  • Read error messages completelyin 89 of 839, across 49 files
  • Create a failing test case before fixingin 84 of 839, across 46 files
  • Reproduce the issue consistentlyin 82 of 839, across 41 files
  • Change one variable at a timein 82 of 839, across 42 files
  • Check recent changesin 74 of 839, across 36 files
  • Write the regression test before fixingin 74 of 839, across 40 files
  • Fix the root cause not the symptomin 60 of 839, across 45 files
  • Implement a single fix at a timein 59 of 839, across 20 files
  • Trace data flow backward to the sourcein 50 of 839, across 20 files
  • Remove all debug instrumentationin 49 of 839, across 13 files
  • Form a single hypothesisin 48 of 839, across 18 files

Said here and by no other author read

  • map candidate code paths from user trigger to symptom
  • write a provisional bug card from available evidence
  • record unavailable facts as unknown
  • read the collector reference before running it
  • capture all five collector values
  • instrument boundaries with privacy-reviewed probes

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 326,851. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.