systematic-debugging
Skill2 starsChanged 52 days agoMITView on GitHub
Debug a failure by recording each hypothesis in the CADENCE assumption ledger and refusing to name a root cause while any hypothesis is still open. Use when a test fails, CI goes red, a gate refuses unexpectedly, behavior contradicts the code, a bug is reported, a fix did not work, or you have already tried more than one thing and are guessing. Use it especially under time pressure, when the cause seems obvious, and when you are tempted to fix the symptom you can see.
190 other projects use a version of this.
Adopted by
Most recently taken up by
- metaspartan/cybaratheir own version, first seen Nov 15, 2026
- alexj11324/Mikkatheir own version, first seen Nov 15, 2026
- mdiri-ai/SoduWorktheir own version, first seen Nov 15, 2026
- komluk/scaffoldingtheir own version, first seen Nov 15, 2026
- KbWen/agent-virtual-officetheir own version, first seen Nov 15, 2026
- Wenaixi/dsh-superpowertheir own version, first seen Nov 15, 2026
Used with
Projects with it also use chinese-code-review, chinese-commit-conventions and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
Leodorareluctant259Skillchinese-code-review
中文代码审查规范——在保持专业严谨的同时,用符合国内团队文化的方式给出有效反馈
Also in 8 projects that use this
Leodorareluctant259Skillchinese-commit-conventions
中文 Git 提交规范 — 适配国内团队的 commit message 规范和 changelog 自动化
Also in 8 projects that use this
obraSkilldiagnosing-superpowers
Use when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.
Also in 20 projects that use this
mjunaidcaSkillbuilding-chatgpt-apps
Guides creation of ChatGPT Apps with interactive widgets using OpenAI Apps SDK and MCP servers. Use when building ChatGPT custom apps with visual UI components, embedded widgets, or rich interactive experiences. Covers widget architecture, MCP server setup with FastMCP, response metadata, and Developer Mode configuration. NOT when building standard MCP servers without widgets (use building-mcp-servers skill instead).
Also in 4 projects that use this
mjunaidcaSkillbuilding-chat-interfaces
Build AI chat interfaces with custom backends, authentication, and context injection. Use when integrating chat UI with AI agents, adding auth to chat, injecting user/page context, or implementing httpOnly cookie proxies. Covers ChatKitServer, useChatKit, and MCP auth patterns. NOT when building simple chatbots without persistence or custom agent integration.
Also in 4 projects that use this
- +2 more from Leodorareluctant259
- +5 more from mjunaidca
Other versions190 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 190
- obra/superpowers★ 314ksame name
- NousResearch/hermes-agent★ 264ksame name
- open-metadata/OpenMetadata★ 16ksame name
- jnMetaCode/superpowers-zh★ 8.5ksame name
- vudovn/ag-kit★ 8.2ksame name
- vudovn/antigravity-kit★ 8.2ksame name
- openai/plugins★ 7.8ksame name
- ChrisWiles/claude-code-showcase★ 6.1ksame name
- farm-fe/farm★ 5.6ksame name
- microsoft/FluidFramework★ 4.9ksame name
180 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository5 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- phase-buildSkill
- pr-landSkill
- release-cutSkill
The file
We split this skill into the rules it gives an agent
6 rules. 0 are common, 4 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 177 edited versions we read keep these.
4
Distinctive
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 13
Every hypothesis becomes a row in CADENCE's assumption ledger, tied to one anchor recommendation, and it must be closed validated or …
- Distinctive: line 27
A symptom you can reproduce: If you cannot reproduce it, that is the finding — report it and gather data.
- Distinctive: line 35
You have not already applied a fix: If you have, say so — you are now debugging two things, and the fix may be masking the symptom.
- Very specific: line 42
Every assumption must belong to a recommendation — assumption add requires --rec.
- Distinctive: line 80
Ask what you would see if the hypothesis were false, not what you would see if it were true.
- Distinctive: line 122
If the bug was a false alarm, say so and close it the same way:
“Common” is measured: kept by at least 6 in 10 of the 177 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Systematic debugging (ledger-backed)
Debugging fails in a predictable way: you form one hypothesis, find something consistent with it, and stop. The trail lives in a transcript nobody rereads, so the next person — often you — starts over.
This skill fixes the second problem structurally and the first by discipline.
Every hypothesis becomes a row in CADENCE's assumption ledger, tied to one
anchor recommendation, and it must be closed validated or rejected before
you may name a root cause.
What is and is not enforced. The ledger is real: cadence assumption rows
persist, and the conclusion query below is deterministic. But CADENCE does not
refuse a conclusion the way settle refuses a phase — no gate fires, nothing
exits non-zero if you conclude with hypotheses open. This is a gate you
honor, not one the engine enforces. Its value is that afterwards anyone can
run one command and see whether you honored it. Do not describe this skill, in
a SUMMARY or anywhere else, as CADENCE enforcing debugging discipline.
Preconditions
- A symptom you can reproduce. If you cannot reproduce it, that is the finding — report it and gather data. It is not a licence to start hypothesizing; an unreproducible symptom cannot falsify anything, so every hypothesis you form is unfalsifiable by construction.
- Read the error first, completely. Stack trace, line numbers, exit code. A surprising number of bugs are already named in the text you skipped.
.cadence/is initialized. The ledger commands write there. In a fresh worktree runcadence onboardfirst.- You have not already applied a fix. If you have, say so — you are now debugging two things, and the fix may be masking the symptom.
Pipeline
1. File the anchor
Every assumption must belong to a recommendation — assumption add requires
--rec. File one for the bug before forming any hypothesis:
cadence recommendation add \
--title "bug: <symptom in one line>" \
--summary "<what you observe, and the exact reproduction>" \
--priority medium --readiness needs-evidence \
--area <area> --file <path/to/suspect.ts>Record the reproduction as its first evidence, so the trail starts with a fact rather than a guess:
cadence recommendation evidence add <recId> --note "Reproduction: <exact steps> -> <exact output>"Note the id. Every hypothesis hangs off it, and you advance it at the end.
2. One hypothesis at a time
cadence assumption add --rec <recId> --text "<falsifiable statement>"Falsifiable means you can state, now, an observation that would prove it wrong. "Something is wrong with the cache" is not a hypothesis. "The cache returns a stale entry because the TTL is compared in seconds against a millisecond timestamp" is.
One at a time. Filing three hypotheses and then investigating is how you end up confirming whichever you happened to test first. If you genuinely have three candidates, file the one you can disconfirm fastest, close it, then file the next.
3. Design the observation that would reject it — then run it
Ask what you would see if the hypothesis were false, not what you would see if it were true. Confirmation is cheap and almost everything is consistent with a wrong hypothesis. Then run it and close the row:
cadence assumption reject <assumptionId> # the observation contradicted it
cadence assumption validate <assumptionId> # the rejection test failed to reject itRecord the observation. validate and reject take only an id — the
ledger keeps the verdict but has nowhere to keep the evidence for it
(rec-20260918-001). Put it on the anchor and name the assumption id in the
note. This is a convention, not a schema link; nothing validates the reference:
cadence recommendation evidence add <recId> --note "<assumptionId>: <what you ran> -> <what you saw> -> rejected"4. The gate — check before concluding
cadence assumption list --filter-rec <recId> --filter-status open --format jsonNon-empty means stop. You have an open hypothesis; either close it or explain in the anchor's evidence why it is no longer relevant and then close it. Do not name a root cause while this query returns anything.
This is the step the whole skill exists to make checkable. Run it as a command and read the output — do not decide from memory that you closed everything.
5. Conclude, and close the anchor
The root cause is the validated assumption whose rejection test failed to reject it. Record it, then advance the anchor so the session leaves no dead row:
cadence recommendation evidence add <recId> --note "Root cause: <validated assumptionId> — <statement>. Fix: <what changes>."
cadence recommendation promote <recId> --status=accepted --readiness=ready-for-cadence-specIf the bug was a false alarm, say so and close it the same way:
cadence recommendation promote <recId> --status=rejectedA rejected anchor with its hypothesis trail intact is a successful debugging
session. "No bug here, and here is what I ruled out" is a real result.
Known failure modes to actively watch
- Concluding on one confirmation. You found something consistent with the hypothesis and stopped. Consistency is not evidence; you must have tried to reject it and failed.
- Batching hypotheses. Three rows filed at once, one investigated, the other two quietly closed to clear the gate. The gate is satisfied and you learned nothing — this is the failure the one-at-a-time rule exists for.
- Closing a row to clear the gate.
validateon an assumption you never tested is the same self-report this repo refuses everywhere else. If you are closing it because you stopped caring,rejectit and say why in the evidence. - Fixing the symptom and closing. The test passes now but no assumption was ever validated. You have a green suite and an unexplained system; the bug will return under a different symptom.
- Leaving the anchor open. Step 5 exists because an abandoned
candidaterow is indistinguishable from real queued work, and the ledger is already large. - Hypothesizing without a reproduction. See Preconditions. If you cannot reproduce it you cannot falsify anything, and the ledger fills with rows that can never be honestly closed.
- Editing the ledger by hand.
.cadence/intelligence/is written through the CLI. A hand-edited row is a fabricated audit trail.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

