Internal shared skill for the technical-analysis stage of code-fixing workflows. Decides when a confirmed problem needs external web research before (or instead of) deeper code-level analysis, and how to act on the results.
Prerequisite check: this skill declares
effective-web-researchin frontmatterdependencies. On load, verify it is available; if missing, abort and print the install command (npx skills add FuDesign2008/open-skills -g). No silent fallback.
Step-number parameterization
This skill is shared by workflows with different step numbering. All jump targets below are written as placeholders — {impact-assessment step}, {root-cause step}, {upstream-eval step} — and the referencing workflow states its own numbering inline at the reference line (e.g. "jump to step 4"). Never hardcode a workflow's step numbers here.
1. Research routing triage
Run immediately after the problem's existence is confirmed. Decide which class the root cause most likely falls into — this adjusts the priority of later steps, never replaces them:
| Route | Signals | Emphasis |
|---|---|---|
| 🟢 Internal-first | About "our code/logic/conventions"; root cause suspected in this repo | Code location first; quick search (§2) as optional fallback; root-cause analysis is the core |
| 🔵 External-first | Named third-party lib/framework/API + version; or "how to use / why / any known issue"; or any §2 trigger hits (platform silent failure / nested host / no code-level suspects) | Quick search (§2) becomes the primary action — search known cases first |
| 🟣 Hybrid (external-then-internal) | External concept + internal object ("does the X lib we use have vulnerability Y", "apply pattern X to our code") | §2 first to understand the concept externally, then internal root-cause analysis to apply it |
When unsure, default to 🟢 internal-first (these workflows exist to change code; most root causes live in the repo).
Relation to
effective-web-researchStep 0: that triage decides whether a single search should go external; this one decides the whole problem's research composition. Complementary — consult its Step 0 when unsure.
2. Known-issue quick search
Triggers (any one; in 🔵/🟣 routing this is the primary action, in 🟢 an optional fallback before instrumenting code):
- Platform/native component involved (Android WebView, iOS UIPickerView, browser API, OS-level control…) with a silent failure — no error, no crash, correct calls, completely unresponsive behavior.
- Nested host runtime (React Native, Electron, Cordova, WebView-in-App…) while the frontend logic looks entirely correct.
- No code-level suspects: call chain complete, parameters correct, logs show execution — yet it doesn't work at runtime.
Execute: WebSearch with "symptom keywords + platform/framework + year" across StackOverflow, GitHub Issues, official docs. Parallel action (mandatory when a named third-party lib/framework is involved or symptoms correlate strongly with platform-specific behavior): check the upstream dependency's Changelog / Release Notes / Issues for an already-fixed version — upgrading may be the simplest fix (hand off to the workflow's {upstream-eval step}, which delegates to upstream-dependency-debug).
Result handling:
| Outcome | Action |
|---|---|
| ✅ Known case found, root cause clear | Output the quick-search conclusion; jump to the {impact-assessment step}, skipping root-cause analysis |
| ✅ Upstream fix version found | Output the lead; proceed to the {upstream-eval step} to assess upgrade feasibility (a fixed-upstream lead is not yet a solution) |
| ⚠️ Related discussion, root cause partly clear | Feed findings into the {root-cause step} as reference |
| ❌ Nothing found | Skip silently; continue to the {root-cause step} |
2.1 Performance-pattern variant (for perf-optimize-workflow)
When the referring workflow is performance analysis (perf-optimize-workflow), treat §2 as a known performance-pattern quick search with these extra triggers (any one, before entering that workflow's hypothesis stage):
- Anomaly located but cannot be classified into the workflow's known performance-pattern table
- Suspected framework/library performance bug or limit (named framework + version)
Execute the same WebSearch discipline as §2, with query shape: symptom + framework/lib + version + year, preferring GitHub Issues, official changelog, StackOverflow.
| Outcome | Action |
|---|---|
| ✅ Known case / version bug | List the known fix as a candidate solution; focus later hypothesis validation on that direction |
| ⚠️ Related discussion, inconclusive | Feed into hypothesis directions |
| ❌ Nothing found | Skip silently; continue the workflow's next stage |
perf-optimize-workflow keeps its pattern table and stage orchestration; this subsection only owns the search methodology.
3. Industry-wide issue evaluation
Trigger (optional, pessimistic branch): the root cause clearly points to a hard platform / language / protocol / standard limit (e.g. browser security policy, JS single-threading, protocol constraints) with no obvious application-layer workaround.
Execute: WebSearch — is this a recognized industry-wide problem? How do mainstream frameworks / major companies handle it? Does any viable workaround exist?
Result handling:
| Outcome | Action |
|---|---|
| 🚫 Recognized hard problem, no viable solution | Output the evaluation report (template in reference.md) and pause for the user's decision (explore workarounds / accept as-is) |
| ⚠️ Limited but workarounds exist | List workarounds as candidates in solution exploration; continue |
| ✅ Not industry-wide, a fixable problem | Continue to the {impact-assessment step} |
Distinction from §2: §2 searches known cases by symptom ahead of root-causing (applies whether or not a solution exists); §3 evaluates after the root cause is clear whether the industry has a recognized solution (for the no-solution scenario). They never overlap.
WebSearch discipline
All WebSearch execution in this skill follows effective-web-research: first its Step 0 triage (confirm the question is external, not solvable from the repo), then its 4 maxims (authority-first / currency check / cross-validate non-trivial claims / skip content farms); when the user demands rigor, switch to its strict mode and produce a report. Read that skill's current doc before searching — never from memory.
Integration guide (for referencing workflows)
- Declare this skill in frontmatter
dependencies; abort at startup if missing. - Reference line must state: your step-number mapping for the placeholders above, and which of your steps §2/§3 slot into.
- Keep your semantic variants in your own body — e.g. jira-fix-workflow: §3 is a gate (not optional), a 🚫 outcome stops the flow and writes a Jira comment, and its report template lives in its own reference.md.