diagnose
Diagnose a reproducible failure, a live runtime symptom, or a supplied trace or profile; fix and verify the cause when requested.
Skill by Yeshwanthyk
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/yeshwanthyk-diagnose
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md and code. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/diagnose
What's inside
- Skill name
diagnose- Description
- Diagnose a reproducible failure, a live runtime symptom, or a supplied trace or profile; fix and verify the cause when requested.
- License
- MIT
- Instructions
- 39 lines in SKILL.md · view on GitHub
Files (3)
- SKILL.md
- runtime-forensics.md · references/
- trace-forensics.md · references/
SKILL.md (first lines)
# Debug Build a tight red-to-green loop. Reproduce the symptom on the matching surface, follow evidence to the first divergence, make the smallest coherent correction, and prove it there. ## Reproduce Pick the starting point first: - A supplied trace, profile, or dump: follow [trace forensics](references/trace-forensics.md); the artifact replaces a live reproduction. - A live runtime symptom to diagnose without a fix: follow [runtime forensics](references/runtime-forensics.md). - Otherwise, reproduce here. When the failure reproduces, read [fix-root-causes](../yesh-router/references/principles/fix-root-causes.md) before tracing it. State the observed symptom and expected contract. Drive the relevant browser, UI, CLI, API, test, or runtime surface yourself. Capture a repeatable red signal. If the surface is inaccessible, record the limitation and the strongest substitute; do not call it a reproduction. ## Find the divergence Trace the failing path through callers, contracts, transformations, state owners, persistence, external effects, and failure propagation. Treat caches, configuration, locks, serialized data, restart behavior, and concurrent writers as suspects when they can change the result. When the correction adds or changes a cache, cleanup, or a concurrent writer, read [state-ownership](../yesh-router/references/principles/state-ownership.md). Form competing hypotheses and choose probes that remove uncertainty. Use focused logging or instrumentation when needed, then remove probes that did not support the surviving explanation. Compare expected and observed values at meaningful boundaries. Find the first divergence rather than the last visible symptom. Check sibling callers or equivalent state paths only when evidence suggests the mechanism is shared. [`how`](../how/SKILL.md) for current mechanics and how's [History](../how/SKILL.md#history) view for regression history are optional aids. One mechanism must explain how the symptom occurs. A plausible story without a confirming probe remains inconclusive. ## Correct A diagnosis-only request ends with the cause and evidence. When a fix is requested, change the first evidenced divergence. Consult [`architect`](../architect/SKILL.md) when the target shape is unsettled. When a cheap local test can pin the red signal, write it before the fix and run it: it must fail for the intended reason, not a setup or unrelated error. Prefer no new test over a bad one: a test of mocks, implementation details, timing, or expensive infrastructure for a small fix. Then use the closest executable check instead (a script, reproduction command, or browser drive). Never edit or weaken a test to match a wrong implementation. Keep validation at boundaries and migrate evidenced sibling callers within scope. Do not add speculative guards, retries, or compatibility branches. When a second fix built on the same assumption fails the same check, stop patching and follow the repeated-fix branch of [fix-root-causes](../yesh-router/references/principles/fix-root-causes.md). Delegation is optional. If useful, follow the [shared delegation contract](../yesh-router/references/delegation.md), assign a precise read-only diagnosis or disjoint edit, then inspect the actual diff and proof yourself. ## Prove and hand off Before calling the fix done, read [prove-it-works](../yesh-router/references/principles/prove-it-works.md). Rerun the original red signal on the same boundary. Add focused sibling, edge, retry/replay, restart, or failure checks when they are part of the mechanism. Inspect the changed diff and remove temporary probes. If the original surface cannot be exercised, report the result as inconclusive. Report the root cause first: red signal, first divergence, causal chain, correction, failing-then-passing evidence, checks run, and remaining uncertainty. Do not imply broader coverage or publication than occurred. ## Completion Diagnosis is complete when the cause and evidence limits are clear. A requested fix also needs a correction at the first divergence and proof on the original boundary plus relevant mechanism-specific checks.
From Yeshwanthyk/agent-skills, MIT.
More from Yeshwanthyk/agent-skills
- 0.627score
plan
Yeshwanthyk
Turn a settled approach into an implementation plan of ordered chunks, dependencies, risks, and proof.
SkillAFreshness: A★ 6Claude Code +9 - 0.627score
motion-explainer
Yeshwanthyk
Build a narrated, timeline-driven systems explainer (boxes, wires, packets, dotted orbs, callouts, code diffs) as a playable HTML explorer, optionally…
SkillAFreshness: A★ 6Claude Code +9 - 0.624score
deep-dive-explainer
Yeshwanthyk
Research a topic and write a long-form, connected explainer as a standalone HTML article, with illustrated or animated explanations when they help teach the…
SkillAFreshness: A★ 6Claude Code +9 - 0.624score
how
Yeshwanthyk
Explain how code works or why it was built this way: execution paths, ownership, inventories, comparisons, design critique, or decision history from commits…
SkillAFreshness: A★ 6Claude Code +9 - 0.624score
interactive-explainer
Yeshwanthyk
Build a source-grounded interactive HTML explanation of behavior or state.
SkillAFreshness: A★ 6Claude Code +9 - 0.624score
invariants
Yeshwanthyk
Model state and break invariants: authority, transitions, concurrency, recovery, with TLA+ or Quint model checking, Lean proofs, or Kani-style verifiers on…
SkillAFreshness: A★ 6Claude Code +9 - 0.624score
meat-review
Yeshwanthyk
Review a diff, commit, branch, PR, or uncommitted change as an abridged reading of what to read, why, and the risks, in chat, a T3 page, or kvim.
SkillAFreshness: A★ 6Claude Code +9 - 0.624score
reflect
Yeshwanthyk
On request, extract reusable lessons from a session and propose or apply scoped instruction changes, or create or update a personal mode from stated…
SkillAFreshness: A★ 6Claude Code +9 - 0.624score
verify-evals
Yeshwanthyk
Verify skill changes in this repo through the skilleval CLI, running only the eval cases added or changed since a base ref and checking that they target the…
SkillAFreshness: A★ 6Claude Code +9 - 0.624score
show-me
Yeshwanthyk
Draw the current topic when asked to see it (show me, draw, diagram, sketch): a viz, tree, diff, or pseudocode.
SkillAFreshness: A★ 6Claude Code +9 - 0.622score
benchmark-checklist
Yeshwanthyk
Vet a performance number before reporting or acting on it.
SkillAFreshness: A★ 6Claude Code +9 - 0.621score
architect
Yeshwanthyk
Choose an architecture when a consequential design boundary is unsettled.
SkillAFreshness: A★ 6Claude Code +9
Signals
- Stars
- 6
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Oct 13, 2026 (17 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/Yeshwanthyk/agent-skills
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from Yeshwanthyk
- 0.627score
plan
Yeshwanthyk
Turn a settled approach into an implementation plan of ordered chunks, dependencies, risks, and proof.
SkillAFreshness: A★ 6Claude Code +9 - 0.627score
motion-explainer
Yeshwanthyk
Build a narrated, timeline-driven systems explainer (boxes, wires, packets, dotted orbs, callouts, code diffs) as a playable HTML explorer, optionally…
SkillAFreshness: A★ 6Claude Code +9 - 0.624score
invariants
Yeshwanthyk
Model state and break invariants: authority, transitions, concurrency, recovery, with TLA+ or Quint model checking, Lean proofs, or Kani-style verifiers on…
SkillAFreshness: A★ 6Claude Code +9 - 0.624score
how
Yeshwanthyk
Explain how code works or why it was built this way: execution paths, ownership, inventories, comparisons, design critique, or decision history from commits…
SkillAFreshness: A★ 6Claude Code +9 - 0.624score
interactive-explainer
Yeshwanthyk
Build a source-grounded interactive HTML explanation of behavior or state.
SkillAFreshness: A★ 6Claude Code +9 - 0.624score
deep-dive-explainer
Yeshwanthyk
Research a topic and write a long-form, connected explainer as a standalone HTML article, with illustrated or animated explanations when they help teach the…
SkillAFreshness: A★ 6Claude Code +9
Cite this page
diagnose skill. Agent Mart. https://agentmart.dev/skills/yeshwanthyk-diagnose (last updated 2026-10-24)