메뉴
Quick debugging patterns and known production failure traps. Reference guide for common issues.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Amend a published CLI from one of two input sources: (1) dogfood mode mines the active Claude Code session transcript for friction (missing flags, hand- rolled API payloads, silent-null returns); (2) direct-input mode accepts user-supplied asks (rename a command, add commands or feeds, fix a named bug, optionally sniff the source site for new endpoints). Confirms scope with the user, plans + executes the fix autonomously, scrubs PII, and opens a PR against mvanhorn/printing-press-library. Two user-in-loop checkpoints: scope after capture, PR draft before open. Trigger phrases: "amend the CLI", "submit a patch", "fix what I just dogfooded", "open a PR for this CLI", "patch this CLI", "add features to my CLI", "rename this command", "add these feeds to <cli>", "sniff for new APIs in <cli>", "amend with these ideas", "use printing-press-amend", "run printing-press-amend".
Bring a published CLI from the public library into the internal library so it's identical to a freshly-generated copy — module path reverted, manuscripts placed alongside, ready for /printing-press-polish or /printing-press-emboss. Use when the public library has a CLI you don't have locally, or to recover from a broken/lost internal copy. Trigger phrases: "import the CLI", "bring it into my library", "fetch from public library", "I don't have it locally yet".
Internal sub-skill: agentic review of a printed CLI's sampled command output for plausibility issues that rule-based checks can't encode (substring-match relevance, format bugs, silent source drops, ranking failures). Invoked via the Skill tool by main printing-press SKILL.md (Phase 4.85) and printing-press-polish SKILL.md during the diagnostic loop. Not for direct user invocation — its actionable wrappers are /printing-press and /printing-press-polish.
Polish a generated CLI to pass verification and become publish-ready. Runs diagnostics (dogfood, verify, scorecard, go vet, gosec), automatically fixes all issues (verify failures, static-analysis findings, dead code, descriptions, README, MCP tool quality), reports the before/after delta, and offers to publish. Use after any /printing-press run, or on any CLI in $PRESS_LIBRARY/. Trigger phrases: "polish", "improve the CLI", "fix verify", "make it publish-ready", "clean up the CLI", "get this ready to ship".
Regenerate an existing printed CLI from scratch under the current Printing Press, with prior research, prior novel features, and prior patches (post-publish hand-fixes) carried into the writing pipeline as reconciliation context rather than dropped on the floor. Pulls the CLI from the public library if it isn't local, recommends reuse-vs-redo of prior research based on age, then hands off to /printing-press with the right context. Use when a machine upgrade would benefit a published CLI more than manual polish. Trigger phrases: "reprint <api>", "regenerate <api>", "redo the <api> CLI", "rebuild <api> from scratch", "this CLI would benefit from a reprint".
Score a generated CLI against the Steinberger bar, compare two CLIs side-by-side
| name | debug |
| user-invocable | true |
| description | Quick debugging patterns and known production failure traps. Reference guide for common issues. |
| allowed-tools | ["Bash","Read","Edit","Write","Grep","Glob","Skill"] |
| model | inherit |
Fast reference for known production failure patterns. For deep debugging, use /carmack.
When the symptom involves a third-party integration falling back to a worse path (handoff, redirect, "manual step required") — Verint dform, Clerk, Stripe, SeeClickFix, OAuth, any SaaS — load ~/.claude/skills/shared/upstream-protocol-investigation.md BEFORE proposing a fix. Read the upstream's primary client source, capture real network traffic, inspect rendered data-* attributes. Treat any "Verified YYYY-MM-DD" comment in our codebase as a hypothesis to re-verify, not a fact. Token cost is unlimited for this; user explicitly authorized it (2026-05-09 SF311 graffiti incident: bandaid 0b746a4 → real fix f27d1e3 after reading dform's api.js lines 462–520).
When each fix makes the symptom change instead of disappear — fetch_error → handshake_rejected → publicid_timeout, every "fix" peeling back to reveal a new failure one layer deeper — STOP. That staircase is the tell that the whole feature premise is wrong, not that you are three fixes from done. Before fixing layer N+1, verify the premise itself: does the real client even use this code path? Capture live traffic and confirm. Reference incident: the SF311 cable chain — four green-tested fixes (wss://→https://, missing Origin header, subscription identifier, publicId timeout) each surfaced the next layer; a live mitm capture then proved the real app never opens the WebSocket at all — the whole path was dead code. Many round-trips of "fix the next layer" should have been one round-trip of "verify the premise."
PREMISE-CHECK GATE — run BEFORE the first fix (not just on the staircase). Full rule: ~/.claude/skills/shared/premise-check.md. Two questions, answered against LIVE upstream docs (not a cached /skill note): (1) Is this approach even valid for THIS runtime/SDK/platform? The browser SDK ≠ native SDK ≠ server SDK — a strategy/option/API documented under one is routinely absent or forbidden in another. If every doc/example for the thing you want sits under a different platform than yours, that's your answer — stop. (2) What's the cheapest probe (curl the API / grep the installed bundle / read the doc's "supported platforms" line) that proves it's possible here? Run it before coding. Docs-before-note (always): a /skill recipe, comment, memory, or prior conclusion is a HYPOTHESIS — re-verify any load-bearing "API/SDK can/can't do X" claim against the upstream's own current docs/source before building on it; live source wins, fix the stale note in the same pass. 2026-06-13 reference incident: hours spent debugging Clerk native oauth_token_apple from a Capacitor webview — clerk-js (browser SDK) can never send it (browser-forced Origin vs Clerk Native-API Authorization conflict), a fact Clerk's docs state plainly under "Expo only." A wrong /ios trap-#4 note was trusted as fact; the working web-OAuth fix was a 5-minute live-doc read away. Now also enforced session-wide by the premise-check-session-start.sh SessionStart hook.
A "fixed" symptom is one that fails to reproduce on the deployed/built artifact, not in the source. After EVERY fix, before declaring root cause resolved, run the No-Lie Verification Protocol from ~/.claude/skills/shared/no-lie-verification.md Checks 3 + 4:
curl https://prod-url/path for the OLD string. Expected: 0 matches.curl -X POST reproducing the original request with a fresh cache-buster. Expected: 200 or the deliberate fallback, never a regression.tools/repro/*.sh) under load and confirm the failure no longer reproduces over N runs (N≥10).catch branch, conditional cron) — induce the trigger and watch the path fire (Check 6 in no-lie-verification.md). Don't infer it from "the parts work": force the 429 / fail the dependency N times / feed the exact bad input on an isolated copy, confirm the fallback/retry/error-path actually executed, then confirm the live instance is untouched. "Configured" ≠ "fires."Reference incident (2026-05-18): /carmack agent reported "no stale '10,000' strings — rg clean" but never curl'd the live URL. Old build was still serving until /ship deployed. The fix wasn't a fix until the deployed artifact was re-tested.
/debug [pattern name or symptom]
/debug catch-all -- Catch-all error handling masking root cause/debug react undefined -- React "X is not defined" scope bug/debug silent startup -- React silently fails to mount/debug auth failed -- Generic auth error hiding real cause/debug broken icons -- Third-party icons/images missing (CSP blocking)/debug text overflow -- Text escaping card boundaries on mobile/debug stale data -- Admin changes not visible to users/debug deploy logout -- Users logged out after every deploy/debug cloudflare security alert -- Cloudflare flagged site for missing security.txt / HSTS / CAA / DNSSEC/debug slow page -- Page/route slow: full-table scan on the hot path, cron warming the wrong cache, or a lagged-source window returning empty/debug cookies -- Need cookies for curl/yt-dlp/scrape from a logged-in browser session (see "Cookie extraction" below)Pattern numbers are globally unique across all reference files, not per-file. Before adding
## Pattern N:, claim the next free N:
grep -rhoE "^## Pattern [0-9]+:" ~/.claude/skills/debug/references/*.md \
| grep -oE "[0-9]+" | sort -n | tail -1 # highest in use
grep -rhoE "^## Pattern [0-9]+:" ~/.claude/skills/debug/references/*.md \
| grep -oE "[0-9]+" | sort -n | uniq -d # existing collisions
#12 and #15 are legacy collisions (two distinct patterns each). Always cite a pattern as
file + number (error-handling-patterns.md #29), never a bare number, so a collision can never
misroute an agent.
Match the user's symptom to the right reference file, then load ONLY that file.
| Symptom / Keyword | Pattern | Reference File |
|---|---|---|
iOS app, simulator, Xcode, Swift crash, .ips crash log, Capacitor shell misbehaving, WKWebView rendering wrong (e.g. border-radius not clipping a composited img), in-app sign-in bounced to Safari, app stuck on splash / OTA rollback, webview UA flagged as in-app browser, Clerk/Apple/Google social sign-in failing in a Capacitor app — authorization_invalid / native_api_disabled / origin_authorization_headers_conflict / oauth_token_apple / "native social login won't work in the app", OR a passkey/WebAuthn ceremony failing in the webview — "passkey registration was cancelled or timed out", webcredentials, associated domains, AASA 404 | iOS App Symptoms — route to the /ios skill (dev/debug loop, axiom crash/build/perf agents, webview driving via axe, simulator streaming/eyes + taps + :3100/ax a11y tree via the /serve-sim skill, AIVA Capacitor traps incl. clerk-js allowedRedirectProtocols, notifyAppReady rollback, WKWebView clip bug). Any in-webview auth/credential ceremony fails with a GENERIC error until the app↔domain binding exists — read /ios trap #12's binding matrix and PROBE FIRST: social-OAuth Apple → trap #4 (allowNavigation); Google → policy-blocked, hide it; clerk-js scheme → allowedRedirectProtocols; passkey "cancelled or timed out" → curl <domain>/.well-known/apple-app-site-association + grep entitlements for webcredentials (improvebayarea dadfdf3). Don't debug the JS ceremony before the 30-sec probe. | /ios skill (~/.claude/skills/ios/SKILL.md) |
| catch-all, generic error, wrong status code, misleading error | Catch-All Error Masking (#1) | error-handling-patterns.md |
| SF311 save 500, Verint save 500, Improve AI resubmit, category changed 500, request_type_id 500 | External Municipal Form Category Hard-500 (#15) | error-handling-patterns.md |
| ticket description truncated, navFooter dropped, map links missing, address missing on filed ticket, description ends abruptly, external form truncation, Verint dform Request_description, bracket truncation, square bracket strip | External Form Description Truncation by Character (#19) | error-handling-patterns.md |
| 311 ticket missing Location box, Open311 address null, mobile311 viewer shows description but no Location dt/dd, "one ticket has location another doesn't", sf_full_address dropped, Location_description ignored, structured location empty, lat/long null in Open311, address only in description body not the actual location field, scf location_details, address forwarding broken, coord-string in structured slot, "hardening 311 address forwarding", multi-city address never break | 311 Structured-Location Dropped by Long-Form / Coord-String Address (#21, backend-agnostic) | error-handling-patterns.md |
| DBI complaint reported as validation but city has it, "DBI re-rendered the form (validation rejected the submission)" but DataSF shows the case, third-party form rejected but actually recorded, form echoed back on success, parser says failure but the agency received it, external form 200 OK we classified as failure, complaintNumber undefined but submission worked, dform/Verint/aspx false-negative on success, success-detection heuristic matches both success AND failure pages, lblError success vs failure span, signal-extraction tests use synthetic HTML, every IBA-submitted DBI complaint fails | Third-party signal-extractor false-negative due to synthetic fixtures (#23) | error-handling-patterns.md + ~/.claude/skills/shared/third-party-signal-fixtures.md |
| CI false positive, grep wrong, CI still fails | CI False Positives (#11) | error-handling-patterns.md |
| admin 403, requireAdmin, metadata-only | Admin Auth Missing DB Fallback (#14) | error-handling-patterns.md |
| admin 500 instead of 403 | Admin Route Wrong Status | error-handling-patterns.md |
| react undefined, scope bug, not defined | React Scope Bug (#2) | react-patterns.md |
| silent startup, blank page, no console errors, module-level throw | Silent React Startup (#3) | react-patterns.md |
| useEffect, renders twice, state lags, derived state | useEffect Abuse (#15) | react-patterns.md |
| localStorage, preference lost, resets on refresh | Preference Lost on Reload (#13) | react-patterns.md |
| double click, duplicate API call, async button | Async Button Double-Click | react-patterns.md |
chat/support widget gone after SSR, cookie banner missing on SSR page, ?support=open/deep-link does nothing, global widget vanished after Hono/Astro/RSC conversion, "worked on every page before SSR", floating launcher dead, analytics/exit-intent dropped on SSR route | Global App.tsx Component Vanishes After SPA→SSR Conversion (#24) | react-patterns.md |
invisible text after SSR, white headings on light/gray bg, page background wrong color after conversion, "colors got messed up on conversion", body background overridden, SSR design clobbered by Tailwind/island CSS, text disappeared but HTML is there, computed bg ≠ design token, low contrast only on SSR pages, island/global CSS leaks body styles | Bundled Island/Global CSS Clobbers SSR Inline Design — Invisible Text (#25) | react-patterns.md |
| auth guard, unauthenticated error, no redirect | Missing Frontend Auth Guard | react-patterns.md |
renders undefined, shows NaN, "Invalid Date", "[object Object]", toggle/section/control disappeared or silently vanished, "make sure nothing is undefined", Cannot read properties of undefined, is not iterable, null-gate hides UI, useState<T|null> gates a render, .map/property access on possibly-undefined API data, admin/dashboard renders garbage, optional chaining missing | Undefined / Null-Render Safety (9-pattern catalog + live DOM grep for undefined/NaN/[object Object]) — also re-sweeps the adjacent admin blind-spot class (default-LIMIT truncation, NULL-aggregate sort burial, count-source mismatch, inner-JOIN row drop, admin-auth DB fallback) | ~/.claude/skills/shared/undefined-null-render-safety.md |
generic isRecord/isObject guard, as unknown as T, (x as any).field, repetitive type-guard boilerplate, "vibe coding" / AI-slop TypeScript, loose unknown-everywhere, no schema validation at boundary, Record<string, unknown> everywhere | Anti-Slop TypeScript — replace generic guards/casts with named types, discriminated unions, or Zod (z.infer); detector: ~/.claude/skills/carmack/tools/detect-ts-slop.sh | ~/.claude/skills/shared/anti-slop-typescript.md |
| text overflow, min-w-0, flex escape, card boundary | Text Overflow Flex+Grid (#12) | css-layout-patterns.md |
| grid mobile, grid-cols, responsive breakpoint | Fixed Grid Breaks Mobile | css-layout-patterns.md |
| iOS Safari, blank iPhone, PDF iframe, vh units | iOS Safari Rendering | css-layout-patterns.md |
| og image, twitter card, social cache, stale card | Social Card Cache (#5) | csp-cache-patterns.md |
| CSP, embed blank, frame-src, img-src, broken icons | CSP Blocking (#9) | csp-cache-patterns.md |
YouTube "Error 153", "Video player configuration error", embed shows gray panel / "Watch video on YouTube", Vimeo/Spotify/SoundCloud embed won't load, video plays on youtube.com but not on my site, embed broke after adding security headers / secureHeaders() / Hono | Embed Dies With No Referer (#27) — NOT CSP, NOT the video. The page sends Referrer-Policy: no-referrer (Hono secureHeaders() default; also .htaccess / WP security plugins / ad blockers). YouTube has refused Referer-less embeds since late 2025. Probe FIRST, 30s: curl -sI https://<site>/ | grep -i referrer-policy (→ no-referrer = confirmed) and curl -s -o /dev/null -w "%{http_code}" "https://www.youtube.com/oembed?url=https://www.youtube.com/watch?v=<ID>&format=json" (200 = video is fine; 401 = embed disabled; 404 = dead). Fix = referrerpolicy="strict-origin-when-cross-origin" on the iframe — never weaken the global header. Verify on the DEPLOYED artifact (hono/jsx vs React attribute casing can drop it) + a real browser screenshot; allow >30s for Worker propagation before concluding. Sweep the repo for the sibling bugs: rendered hrefs containing whitespace (renderer swallowed prose into the URL) and bare URLs rendering as dead text. | csp-cache-patterns.md (Pattern 27) |
| Cloudflare security alert, Security Insights CSV, security.txt missing, /.well-known/security.txt 404, securityheaders.com low score, observatory grade D, missing HSTS/CSP/COOP/CORP, CAA records, DNSSEC, RFC 9116, sitemap.xml 404, robots.txt advertises sitemap that doesn't exist | Site Security Baseline (#16) | ~/.claude/skills/shared/site-security-defaults.md |
CF Security Insights flags: "Dangling A/AAAA", "Unproxied CNAME", "DMARC Record Error", "Bot Fight Mode not enabled" — DO NOT blindly apply CF's recommended fix; ~16% are false positives that break Clerk/SaaS CNAMEs, delete live Google-forwarding DNS, or duplicate valid DMARC. dig/curl-verify the OFTEN-FALSE classes first. Apply only the 3 safe fixes (security.txt + block AI bots + AI Labyrinth). | CF Security Insights triage map | ~/.claude/skills/shared/site-security-defaults.md (triage section) + ~/.claude/skills/carmack/tools/cf-security-insights.sh |
DMARC failing from a sender IP, custom-domain mail bouncing / landing in spam, "send-as alias DMARC fail", Gmail "Send mail as", p=reject rejecting my OWN mail, "SPF passes but DMARC fails", "can't send from my domain", forwarded mail failing DMARC, a DMARC aggregate (RUA) report flags a source — this is OUTBOUND auth alignment, NOT broken records. Records are usually fine; the sending path is wrong (Gmail send-as / a relay signs as the wrong domain → no aligned identifier → DMARC fail). dig the real SPF/DMARC/DKIM/MX FIRST; adding the relay's IP to SPF usually does NOTHING (SPF aligns on the envelope domain). Check the platform's CURRENT native send capability before recommending a 3rd-party — verify live, not from memory: a Cloudflare domain can now SEND via Email Sending (wrangler email sending list/settings; smtp.mx.cloudflare.net:465, user api_token, pwd = CF token w/ Email Sending Write). Fix = route the From-domain's sending through a DKIM-aligned sender; verify with a REAL send (gog send --from <alias> → grep dmarc=pass). | Email Deliverability / DMARC Alignment | ~/.claude/skills/shared/email-deliverability-dmarc.md |
provider swap, replace Resend/Auth0/Stripe/S3/Twilio with X, migrate email/auth/payments provider, "switch from X to Y", removed the old SDK, deleted the old API key, feature silently stopped after swapping providers, if (!env.OLD_KEY) gate, legacy ids unresolvable, old provider ids in DB | Provider Migration Safety — four silent breakages, none of which throw: (1) feature gates still testing the OLD provider's env var (incl. CRITICAL_ENV_VARS) → feature disabled forever once the secret is deleted; (2) persisted legacy identifiers the new API can't resolve → zero rows → a confident wrong status; (3) the new SDK returns {data,error} instead of throwing → every failure reads as success; (4) the local emulator (wrangler dev) doesn't enforce the new provider's server-side rules → local green ≠ prod accepted. Run the 7-item checklist. Never decommission the old account off one repo's grep — another project may hold a live key on the same verified domain. | ~/.claude/skills/shared/provider-migration-safety.md |
status chip says "expired"/"not found" on rows that are fine, freshly-created record reads as missing, a uniform block of rows all show the same scary status, status derived from analytics/GraphQL/search backend, zero rows treated as a reason, retention_expired on brand-new data | Absent Data Reported As A Confident Wrong State (#29) — rows.length === 0 has ≥4 causes (ingestion lag, wrong id namespace/provider, real retention expiry, never created, query failed) and they need different words. Discriminate on identifier SHAPE before querying (free), use the system of record's created_at vs the backend's ingestion grace window to separate lag from expiry, model reasons as a union, and never render a raw enum in the UI. | error-handling-patterns.md (Pattern 29) |
third-party submit "worked" (200 + id, no error) but the ticket/record never went live — never got a public number, never opened, never associated, openedAt: null, publicId: null, submittedTickets: 0; guest/anonymous create dead-ends silently; A/B-tested a flag/category/field and the symptom stayed IDENTICAL; "why does spotmobile/guest submit never get a case number"; freshly-minted guest token; shadow-limited actor; "is it the payload or something else" | False-Positive Success — async lifecycle is the real signal + IDENTITY is the hidden variable (#30) — a synchronous 200+id is a pending state, not done; gate success on the lifecycle transition (public number / opened / row in public dataset), polled, against a known-good baseline. When fix-A/fix-B/fix-C leave the symptom byte-identical, STOP editing the request — the actor/identity/account you held constant is the variable; re-run the SAME payload under a known-good established identity FIRST. Prove the working backend by RUNNING THE REAL FUNCTION LIVE (repro harness → real id), never from old code or a "Verified" comment. | error-handling-patterns.md (Pattern 30) |
"the deploy broke prod", Failed to load module script ... MIME type "text/html", SPA won't mount after deploy, only SEO fallback text renders, ERR_BLOCKED_BY_CLIENT, page worked 5 min ago, verifying a deploy in a :9222/clone browser | Your Verification Browser Is Stale — False "Prod Is Broken" (#28) — a cached HTML shell references pre-deploy asset hashes; the SPA fallback serves index.html for the missing .js; the browser refuses HTML as a module script. Production is fine; only your tab is broken. NEVER conclude a bad deploy from a browser alone: curl the live HTML for its asset refs, confirm each returns 200 text/javascript (not text/html), and grep the deployed chunk for a string only your new code has. location.reload(true) is a no-op; use navigate_page {ignoreCache:true} or a cache-busting query param. ERR_BLOCKED_BY_CLIENT = ad blocker, not your code. | csp-cache-patterns.md (Pattern 28) |
npm install fails with "Socket npm exiting due to risks", socket blocks install, SOCKET_CLI_VIEW_ALL_RISKS=1 prints nothing, "socket won't let me install", install works with SOCKET_CLI_ACCEPT_RISKS=1 but not without, deps missing after a "successful" npm ci, Cannot find type definition file for '@cloudflare/workers-types' right after install, vitest/tsc missing though they're in package.json | Socket CLI fails CLOSED when it has no API token (verified 2026-07-23) — it aborts the install AND cannot name the risk, because naming it requires the API the missing token would authenticate. The failure LOOKS like a security finding; it is a config gap. Diagnose in one command: ~/.claude/skills/shared/tools/socket-health-check.sh --live (exit 1 = BLIND, 2 = DISARMED, 0 = OK). ⚠️ The install is only PARTIAL — node_modules may be half-populated, so the next symptom is a missing type package or a missing test runner, which sends you debugging the wrong thing. Fix the token, then re-run a full install. Do NOT reach for SOCKET_CLI_ACCEPT_RISKS=1 (permanent blanket disarm — the pre-bash-socket-disarm-guard.sh hook blocks persisting it); use ~/tools/npm-real for a scoped bypass. | ~/.claude/skills/shared/tools/socket-health-check.sh |
dependabot alert, npm audit, CVE, GHSA-xxxx, "vulnerable dependency", transitive vuln, "fix the security/dependabot page", overrides, esbuild/tar/minimatch/ws/vite/uuid vuln, npm install fails on override version | Dependency Vulnerability Audit & Fix (7-step: BOTH dependabot-api + per-manifest npm-audit, scan EVERY lockfile, scoped overrides for multi-major packages, verify patched version is published, never blind npm audit fix --force, verify build with project's real build) | ~/.claude/skills/shared/dependency-audit.md |
| www redirect, cookies lost, auth break | WWW Redirect Auth Break (#4) | csp-cache-patterns.md |
| deploy logout, chunk hash, preloadError | Deploy Logs Users Out | csp-cache-patterns.md |
| stale version, failed deploy, CF Pages | CF Pages Stale Deploy | csp-cache-patterns.md |
| terraform destroy, infra deleted, catastrophic | AI Agent Destroys Infra (#6) | infrastructure-patterns.md |
| prod auth, pk_test, accounts.dev, Clerk production, Wrangler binding, wrong auth instance | Production Auth Instance Drift (#12) | infrastructure-patterns.md |
| prime-rhino, clerk.example.com missing, live HTML/CSP differs from dist, prod integration fails | AIVA Live Artifact/Auth Drift (#12a) | infrastructure-patterns.md |
| wrangler secret put binding already in use, plaintext Worker var, secret_text missing, BREVO_API_KEY in vars | Worker Secret Binding Drift (#12b) | infrastructure-patterns.md |
D1 route 500s with generic "Internal Server Error", no such column/no such table, D1_ERROR, intake/any-write save fails for everyone, migration file exists but column missing in prod, wrangler tail shows outcome Ok + empty exceptions, schema drift, migration not applied to remote | D1 Schema Drift — migration-file-exists-but-not-applied-to-prod (#26). Detect: wrangler tail --format json → grep logs for D1_ERROR/no such column (it's in logs[], not exceptions[]); confirm with PRAGMA table_info(<t>) on --remote (the file/local-D1 are NOT proof). Fix: additive ALTER … ADD COLUMN/CREATE … IF NOT EXISTS on remote, then INSERT OR IGNORE the filename into d1_migrations — NEVER bulk migrations apply (re-runs non-idempotent data migrations → corruption). Gate: ~/.claude/skills/shared/tools/d1-schema-drift-check.sh <repo>. | ~/.claude/skills/shared/d1-schema-drift.md |
| rust clippy, cross-platform, cfg, unused_mut | Rust Cross-Platform Lint (#7) | infrastructure-patterns.md |
| homebrew, keg-only, launchd, cron PATH | Homebrew Keg-Only PATH (#8) | infrastructure-patterns.md |
launchd/cron watchdog false alerts, "job-not-loaded" but service is up, gateway DOWN alert spam, bootstrap rc=5, launchctl list empty in launchd context, health-check passes interactively but fails on schedule, watchdog restarting a healthy service | launchd-Context Health-Check False Positive (#19) | infrastructure-patterns.md |
| serde, deny_unknown_fields, config crash | Serde Config Crash (#10) | infrastructure-patterns.md |
| innerHTML, XSS, dangerouslySetInnerHTML | XSS via innerHTML | infrastructure-patterns.md |
| JSON-LD, script breakout | JSON-LD Breakout | infrastructure-patterns.md |
Python SIGABRT / "Abort trap: 6" at finalization, parent sshd-session, _enter_buffered_busy, Py_Finalize/_io_TextIOWrapper_close in the stack, a Thread blocked in read(), recurring Python-*.ips, "crash over SSH", MCP bridge/server crashes on disconnect, cross-device ssh host python …, crash survives swapping the python interpreter | Python finalization SIGABRT over SSH — daemon-thread-blocked-in-read SCRIPT bug, usually a cross-device MCP stdio server (NOT the interpreter; NOT scheduled — on-demand). Forensic capture via ~/.zshenv SSH-cmd logger + WatchPaths crash-watcher correlated by parentPid. Fix: run the MCP server locally where the resource lives + os._exit(0) after the work loop (#22) | infrastructure-patterns.md |
| cannot find module, postinstall, npm install broke, every update breaks | File Extension Mismatch / Broken Postinstall (#17/#18) | infrastructure-patterns.md |
| npm-cli.js, shell script, SyntaxError unexpected string | npm-cli.js Shell Script Corruption (#17) | infrastructure-patterns.md |
| installing deps one by one, each reveals another missing | Non-Fatal Postinstall Cascade (#18) | infrastructure-patterns.md |
| stale data, admin user sync, visibilitychange | Data Stale Admin/User (#16) | debugging-discipline.md |
| slow page, slow route, slow dashboard, 8M rows, COUNT(*) on every request, full table scan, why is X so slow, route reads wrong cache, cron warms wrong KV key, prewarm warms wrong thing, 24h/last-N window empty, lagged data source, requested_datetime lag | Hot-Path Data-Volume & Cache-Topology (#20) | debugging-discipline.md |
| systematic, 4-phase, root cause, discipline | Debugging Discipline | debugging-discipline.md |
| lint, biome, npm audit, security cleanup | Lint & Security Cleanup | debugging-discipline.md |
Workers Cache leak / wrong data after enabling cache.enabled — user A sees user B's data, authed JSON route returns someone else's response, cf-cache-status: HIT on a per-user route, stale per-user data exactly ~2h old, data leak after enabling the new CF Workers Cache, cache flag "not doing anything" (wrangler <4.69 silently ignores it), Unexpected fields found in top-level field: "cache" | Workers Cache heuristic-caching cross-user leak — a 200 with NO Cache-Control is heuristically cached 2h; cookie auth (Clerk __session) does NOT auto-bypass. Fix: fail-safe default private, no-store + explicit public opt-ins; verify with ~/.claude/skills/ship/tools/workers-cache-check.sh <repo> + live curl -sI | grep -i cf-cache-status. Wrangler must be ≥4.69 in the repo's node_modules or the flag is inert. | ~/.claude/skills/shared/workers-cache-safety.md |
Sitewide redirect loop / site down right after enabling Workers Cache — ERR_TOO_MANY_REDIRECTS, every path 301s to its own URL, all UAs affected, Location: equals the requested URL, 301s explode in zone analytics right after a cache-enable deploy, HSTS header disappeared, worker logs show http:// in request.url while cf-visitor says {"scheme":"https"}, "I disabled the flag and it didn't fix it" (checked within seconds) | Workers Cache scheme-presentation class (2026-07-06 exampleapp ~25-min outage): the cache front-layer hands the Worker http:// for HTTPS visitors → any url.protocol === "http:" canonicalize middleware 301-loops the whole site, and url.protocol === "https:"-gated HSTS silently never fires. Diagnose in ONE probe: wrangler tail --format json → event.request.url scheme for a live HTTPS request. Fix: derive the client scheme from cf-visitor (authoritative) — visitorIsHttps() pattern, AIVA src/worker/middleware/securityHeaders.ts; url.protocol only as local-dev fallback. Triage rules: disable/rollback propagates in >30s — a <60s check is NOT evidence (a 3s check false-negatived during the incident and caused an unneeded rollback; verify over ≥60s); zone SSL mode, Page Rules, and Config Rules are red herrings (the presentation comes from the cache layer itself). | ~/.claude/skills/shared/workers-cache-safety.md |
| error too vague to act on, ambiguous error message, "can't tell what failed", same error fires 100s of times, log noise / superfluous log entries, no log at the failure point, silent failure, "why are these errors so unclear", add debug attributes, clean up the logs, assess the logs, improve log output | Observability / Log Hygiene — make ambiguous errors actionable (what was attempted + actual values + likely cause/branch + disambiguation: a vague error is usually two root causes sharing one string — split them), reduce noise by downgrading level (never delete), add the one attribute that names the failure. For a proactive scheduled pass over a window of logs, use the /log-hygiene skill instead. | ~/.claude/skills/shared/observability-instrumentation.md (+ /log-hygiene skill) |
All reference files are in ~/.claude/skills/debug/references/.
| File | Content |
|---|---|
error-handling-patterns.md | Catch-all masking (#1), external municipal form hard-500 (#15), CI false positives (#11), admin auth DB fallback (#14), admin 500 vs 403 |
react-patterns.md | Scope bug (#2), silent startup (#3), useEffect abuse (#15), localStorage persistence (#13), async double-click, missing auth guard |
~/.claude/skills/shared/undefined-null-render-safety.md | Undefined/null-render bug class — 9 patterns (null-gate hides UI, unguarded property access, .map/string/Date on possibly-undefined, JSON.parse, raw-undefined render, missing loading/empty states, backend omits a field) + live DOM grep for rendered undefined/NaN/[object Object]. Cross-refs the admin blind-spot class (truncation, NULL-aggregate sort, count mismatch, inner-JOIN drop, admin-auth DB fallback). Reference incident: AIVA admin "Test account" toggle hidden by an isTest !== null gate (2026-06-01) |
~/.claude/skills/shared/anti-slop-typescript.md | Anti-slop TypeScript — ban generic isRecord/isObject guards, as unknown as T launder-casts, (x as any).field reach-casts; require named types / discriminated unions / Zod schemas (z.infer) at trust boundaries; targeted predicate only as last resort. Detector: ~/.claude/skills/carmack/tools/detect-ts-slop.sh. |
css-layout-patterns.md | Text overflow flex+grid (#12), fixed grid mobile, iOS Safari rendering (PDF, vh, fixed) |
csp-cache-patterns.md | Social card cache (#5), CSP multi-layer blocking (#9), www redirect (#4), deploy chunk invalidation, CF Pages stale deploy |
infrastructure-patterns.md | Infra destruction (#6), Rust cross-platform lint (#7), Homebrew keg-only (#8), serde config crash (#10), XSS innerHTML, JSON-LD breakout |
debugging-discipline.md | 4-phase workflow, diagnostic checklist (23 items), lint/security cleanup, data stale admin/user (#16), hot-path data-volume & cache-topology (#20) |
~/.claude/skills/shared/observability-instrumentation.md | Observability / log hygiene: 4-part actionable error-message design, boundary/decision-point logging, downgrade-noise-never-delete, instrument-on-fix attribute heuristic, stack log sources. Proactive scheduled counterpart: /log-hygiene skill. |
~/.claude/skills/shared/ant-verification-protocol.md | Ant-level quality gates: OWASP sweep, truthfulness protocol, closed-loop verification |
When you need browser cookies for a non-browser tool (curl --cookie, yt-dlp --cookies, requests cookies={}, scrapers, n8n nodes), pick by where the live session is:
| Where the session is | Best tool | Why |
|---|---|---|
| Live REAL Chrome (fcdp-driven) | ~/tools/fcdp/fcdp raw Network.getCookies '{"urls":[…]}' | Includes session-only cookies not yet flushed to SQLite; no Keychain prompt; aligned with /feedback_browser_default_chrome_profile. |
Regular Chrome profile (~/Library/Application Support/Google/Chrome/Default) | ~/tools/cookies-txt <url> | Reads SQLite + macOS Keychain. Works headless / from cron / without :9222 attached. Outputs Netscape, JSON, or Cookie: header. |
| Brave / Edge / Chromium | ~/tools/cookies-txt --browser brave | edge | chromium <url> | Same code, different Keychain entry. |
| One-shot, don't care about reuse | yt-dlp --cookies-from-browser chrome:Default <url> | Already installed; bypasses the question entirely if all you need is yt-dlp. |
~/tools/cookies-txt reference (ported from the "Get cookies.txt LOCALLY" Chrome extension, source at ~/re/cookies-txt-locally/):
cookies-txt github.com # Netscape → stdout
cookies-txt github.com -o /tmp/cookies.txt # → file
cookies-txt --all -f json # every cookie, JSON
cookies-txt example.com -f header # "name=value; name=value;"
cookies-txt --browser brave --profile "Profile 1" foo.com
Exit codes: 0 ok, 3 no DB at expected path, 4 Keychain entry missing/denied, 5 decrypt failure.
chrome-devtools MCP path for the :9222 Chrome — request CDP Network.getCookies with {urls: ["https://github.com"]}; format the returned cookie array with the same Netscape mapper the extension uses (<domain>\t<TRUE/FALSE includeSubDomain>\t<path>\t<TRUE/FALSE secure>\t<expiry>\t<name>\t<value>).
Don't use unbrowse / agent-browser for cookie extraction — those launch separate browser instances with no logged-in state.
When debugging any production error, check in order:
Is the error message accurate? (catch-all masking?)
Check runtime logs: wrangler tail or Cloudflare dashboard
Reproduce locally with same endpoint + token
Check external services (Clerk/Stripe/DB responding?)
Check env vars in deployment environment
Check www redirect (losing cookies?)
Check third-party IDs (stale template/product IDs?)
Check CSP (all six directives for embeds?)
Text overflow on mobile? (min-w-0 missing, grid-cols-2 without sm:)
Preference not persisting? (localStorage writer without App.tsx reader)
"Cannot find module" after every update? Run file <path>.js — may be a shell script (#17)
Installing missing deps one-by-one, each reveals another? STOP — postinstall is broken (#18)
External city form 500 after category change? Probe the failing category and a known-good category through the same first-party save path before blaming auth/photo.
Production auth configured? Verify Worker dry-run bindings and live HTML do not expose pk_test, sk_test, or .accounts.dev.
Live artifact matches production? Fetch cache-busted HTML and CSP headers; compare live response, not only local dist.
Worker secrets correct? Sensitive bindings must appear as secret_text from wrangler secret list; no key/token/password literal should remain in wrangler.json vars.
wrangler secret put says binding already exists? Remove the plaintext var from config, deploy without --keep-vars, put the secret, and re-list secrets before trusting it.
AIVA deployed? Verify clerk.example.com is present, prime-rhino-99/.clerk.accounts.dev/pk_test/sk_test are absent, and npm run test:integration:prod passes if available.
Page/route slow? (a) Enumerate every query that runs before the response — measure each COUNT(*)/SUM(CASE...)/whole-partition aggregate with wrangler d1 execute --json → meta.rows_read; >~100k rows on a per-request path = the bug. (b) Does it read a cache? Then grep for who writes that exact key — if nothing does (or only a manual script, no cron), it's a permanent MISS. (c) Cron warming a cache? grep for who reads those keys — a warmer that warms keys nothing reads (or a different key-shape than the route reads) is the bug. (d) WHERE ts >= now() - N against a lagged source returns empty — anchor on MAX(ts). See Pattern #20.
Third-party submit reported as failed but the agency received it? (DBI complaint flow, Verint dform save, OAuth callback, any webhook receiver, any reverse-engineered API replay.) Run ls src/__fixtures__/<integration>/ 2>/dev/null. If there is no *-success.* AND *-failure.* pair captured from the LIVE system, you cannot trust the detector — the test suite is asserting against fiction. Write tools/repro/<integration>-probe.{sh,mjs}, run it once for a known-good and a known-bad request, save both responses, then diff them to find the actual distinguishing feature (anchored DOM id / JSON field, NOT a substring that appears in both). Don't ship detector code without real fixtures. See Pattern #23 + ~/.claude/skills/shared/third-party-signal-fixtures.md. Reference incident (2026-05-27): every IBA-submitted DBI complaint was reported as validation error for ~10 days while the city actually recorded every one of them — synthetic test fixture (<html><body><h1>Thank you</h1>...) didn't contain the form layout strings (Sub_Button0, CheckBox1) that the real DBI success page does contain, so the detector's looksLikeEchoedForm heuristic was never exercised against reality.
Error too vague to act on, firing repeatedly, or no log at the failure point? Don't just patch this one instance — make the message actionable (what was attempted + actual values + likely cause/branch + disambiguation; a vague error is usually two root causes sharing one string), add the one attribute that would have named the failure, and downgrade noisy lines rather than deleting them. Standard: ~/.claude/skills/shared/observability-instrumentation.md. For a scheduled sweep over a window of logs, use the /log-hygiene skill.
For the full checklist, load debugging-discipline.md.
/debug is for investigation, local reproduction, local/reviewable fixes, and verification only. Production release authority belongs to /ship, not this skill.
NEVER deploy, promote, release, or mutate production traffic while operating under /debug, even if the user also mentions "ship", "prod", "deploy", "push live", or "fix and deploy" in the same request. Do not run production deployment commands and do not invoke /ship autonomously.
Forbidden examples include wrangler deploy, wrangler versions deploy, wrangler pages deploy, npm run deploy, pnpm deploy, vercel deploy --prod, netlify deploy --prod, firebase deploy, fly deploy, railway up, render deploy, Docker image push/promote commands, production migration/apply commands, and any Cloudflare/Vercel/GitHub release promotion that changes live traffic.
Allowed under /debug: inspect logs, reproduce failures, patch locally, run bounded tests, and commit changes if the user asked for code changes and git safety checks pass. After implementing a fix, STOP and tell the user: "Fix is committed and ready. Run /ship to deploy to production." If the user wants production deployment, they must invoke /ship explicitly in a separate step.
Before proposing any fix that touches a dependency/plugin/framework/third-party widget, load ~/.claude/skills/shared/installed-source-ground-truth.md and follow it: read the installed package's types+JSDoc in node_modules/ for real option semantics, read the native platform implementation when platform support matters, and probe the live DOM/traffic for third-party widgets. Cite file+symbol in the fix. (2026-06-12: three CSS workarounds failed against the Brevo widget; the real fix — StatusBar overlaysWebView:false on iOS — was found by reading the installed plugin's Swift source.)
When a /debug session lands a code fix, run the built-in security-review skill on the fix before declaring it resolved, and loop until it is clean. A debug fix that patches the symptom but introduces (or leaves) an injection / XSS / SSRF / auth-bypass / secret-exposure / logic flaw is not done. This is AI/semantic dataflow analysis — it catches the vulnerability class the pattern-routing tables above do not.
⚠️ OPERATIONAL (verified 2026-07-22): security-review runs against the CURRENT WORKING DIRECTORY's git repo (no path arg; hard-fails "needs to run inside a git repository" if cwd is wrong — cd <repo-root> FIRST) and reviews the COMMITTED branch diff vs the base, NOT uncommitted working-tree edits (proven: unstaged changes → empty diff). Commit the fix before running this gate, or it reviews nothing. It returns a markdown report (file:line, severity, confidence 1–10) with its own false-positive filter at confidence ≥8. If the Skill returns unknown-skill, tell the user the gate couldn't run rather than passing silently.
The loop: (1) cd into the repo root, invoke the security-review skill (Skill tool) on the changed code; (2) triage its reported (confidence-≥8) findings real vs. genuine-false-positive; (3) FIX every real finding inline, obeying the No-Suppression Rule (no @ts-ignore/eslint-disable/biome-ignore — remove the actual vulnerability); (4) re-invoke security-review; (5) repeat until the report is empty. Document any genuine false positive inline with its reason; fix everything else, never defer. Loop guard: same finding surviving 5 attempts → STOP and surface it to the user. This runs entirely within /debug's allowed scope (local fix + verify) — it does NOT deploy (Deployment Prohibition still holds); the code is left security-clean and committed, and /ship's Phase 1.29 re-runs the same loop at deploy time. Skip only when the session made no code change (pure investigation/reference).
Vitest fork workers leak ~5GB memory each when they hang:
timeout 120 npx vitest run src/specific/test.ts 2>&1npm test, npx vitest run with no args)pgrep -f vitest | xargs kill 2>/dev/nullterraform destroy, terraform apply -auto-approve, DROP TABLE/DATABASE, or cloud CLI delete/terminate commandsterraform plan output and get approval before any applyWhen this skill is invoked:
~/.claude/skills/debug/references/~/.claude/skills/shared/ant-verification-protocol.md and apply:
/carmack for deep investigationdebugging-discipline.md)debugging-discipline.md for the commands)Load the matching reference file and show the relevant pattern.
If the issue needs deep investigation beyond known patterns, recommend /carmack.
CRITICAL: Do NOT deploy, promote, release, or invoke /ship from /debug. Production shipping belongs only to /ship, invoked separately by the user.
ssh host 'cmd' (BatchMode), launchd agents, and cron run with a bare PATH that excludes
Homebrew (/opt/homebrew/bin) and ~/.local/bin. The failures below are NOT real — they are PATH:
ssh host 'which node' / which hermes / which brew → "not found" even when installed
(/opt/homebrew/bin/node, ~/.local/bin/hermes). Probe explicit absolute paths; never
conclude "not installed" from which in a non-interactive shell./opt/homebrew/bin/node script.mjs) and set PATH in the plist
EnvironmentVariables. A #!/usr/bin/env node shebang fails under launchd.ssh host 'PATH=$HOME/.local/bin:/opt/homebrew/bin:$PATH <tool> ...'
— a plain ssh host '<tool> ...' false-negatives ("NO-JOB" when the job exists).ssh 'script' on an unmatched glob (/path/*/bin) — quote globs
or wrap in bash -lc.Reference incident (caltrans-pra → mac-mini): ssh mac-mini 'which node' said "not found" (node was
/opt/homebrew/bin/node v26); hermes cron list | grep caltrans said NO-JOB (hermes at
~/.local/bin, off the BatchMode PATH) — the job WAS registered. Both were PATH false-negatives,
not missing artifacts. Verify on the runner via the real invocation path, with an explicit PATH.