SuperAPI AI Directory

web-research

Writing & research

View repository

Updated 10-16 00:21

About this skill

Web Research

Rules

  • Honor user constraints, including no-write/chat-only requests. Discover available retrieval capabilities; do not assume tool names.
  • Prefer primary sources and read answer-critical content. Search snippets and generated summaries are discovery aids, not sufficient evidence.
  • For dependency claims, establish installed/targeted versions and runtime/integration boundaries. Prefer version-matched source over conflicting live documentation.
  • For signatures, hashing, serialization, canonicalization, authentication, or wire protocols, require an independent official implementation, test vector, or interoperability check; self-consistent round trips are insufficient.
  • Cite URLs, separate observation from inference, and state uncertainty. Never store secrets, credentials, private sessions, or sensitive source content.

Resources

Read before the corresponding work:

Work Resource
Substantial research notes or decision synthesis ADR conventions: shared document paths and decision authority
Complex, filtered, multi-angle, or current discovery Web search
Multiple sources, repositories, PDFs, or truncated content Content retrieval
Video/audio or visual moments Media

Use browser automation for logins, forms, or dynamic interaction; local parsers/screenshots for local documents.

Workflow

  1. Frame. Small direct lookups need no artifact. For substantial, conflicting, resumable, implementation-critical, or high-risk research, read ADR conventions and reuse the active work document or create adrs/work/<research-slug>.md. Record question, scope, constraints, known context, and search angles. No-write → retain evidence in the response.
  2. Establish context. Inspect manifests, lockfiles, configuration, or exact source when applicable. Identify the behavior, compatibility, failure, and integration questions that affect the answer.
  3. Discover. Inspect user-supplied sources first; otherwise use two to four distinct angles, such as official docs, source/release history, and corroborating implementations. Retrieve selected answer-critical sources.
  4. Inspect. Read actual docs/source/tests or media. Resolve truncation through full content or an alternate retrieval path. In substantial research, retain useful findings, citations, version/date scope, material conflicts, rejected evidence, open questions, and next action in the same work document.
  5. Resolve. Assess authority, directness, version match, and failure behavior. Follow up narrowly on material gaps; stop when supported, remaining gaps are non-material, or further useful evidence is unlikely.
  6. Synthesize. Reread existing work notes before final responses, handoffs, or compaction. State findings, implications, source conflicts, and limits. Link decision-relevant evidence to the owning ADR; draft only actual significant choices and never treat research as acceptance. Reusable technology knowledge belongs in the wiki.

Late results warrant a follow-up only if they materially change the conclusion or the user asks.