Source Finder

This is an OMH source-finder workflow skill, projected for Agent Skills hosts (Claude Code, Codex, Cursor, opencode, OpenClaw, pi).

Why This Exists

source-finder exists so Hermes can turn vague source discovery requests into typed candidates, acquisition status, and downstream workflow choice without pretending OMH searched, downloaded, or verified the material.

Do Not Use When

• The requested output is factual findings, comparison, or a summary rather than a typed candidate inventory and acquisition status; use research.
• The user needs a business decision brief with evidence-versus-inference treatment; use research-brief.
• The user asks for current citations, fact-finding, or source-backed synthesis; use research.
• The user supplies a paper/PDF/arXiv/DOI/excerpt and wants explanation; use paper-learning.
• The user asks for recurring monitoring, source inbox, or Scout/Analyst/Briefer operations; use research-department.
• The user asks to export, convert, render, package, or attach a file; use materials-package or deliverable-package.
• The user asks for an image card or visual summary; use img-summary.

Examples

Good example:

• Prompt: source-finder find papers, datasets, and GitHub repos for evaluating browser agent benchmarks.
• Expected behavior: Prepare source_finder_plan/v1 with typed candidates, acquisition states, missing observed evidence, and downstream choices.
• Why: The user needs source candidates before deciding whether to learn, research, package, or implement.

Bad example:

• Prompt: source-finder find current citations and summarize what the sources say.
• Expected behavior: Route to research because the user asks for current evidence and synthesis, not candidate acquisition status.
• Why: Source-finder prepares acquisition lifecycle metadata; research owns current evidence synthesis.

Completion Checklist

• Source kinds, source boundaries, and downstream intent are named.
• Each candidate has a source_candidate/v1 shape and acquisition state.
• Observed states include provenance before being treated as evidence.
• The next downstream workflow is recommended without claiming it ran.
• Search, download, clone, extraction, hash, license, verification, and downstream processing gaps are explicit.

Recovery Notes

• If the user asks for facts or citations, route to research.
• If a candidate lacks a link or file reference, keep it candidate_prepared and ask for the next observable source step.
• If the user wants to process a selected source, route to the downstream workflow instead of continuing source acquisition.

Use When

Use when the requested output is a typed source candidate inventory and acquisition status across papers, web links, datasets, GitHub repositories, public presentations, docs/specs, or unknown source material before choosing paper-learning, research, research-brief, research-department, materials-package, or an ultrawork delivery cycle.

Strong routing signals: `source-finder`, `source finder`, `source acquisition`, `source intake`, `find papers and datasets`, `find datasets and repos`, `find papers`, `find arxiv link`, `find arxiv paper`, `find datasets`, `find github repos`, `find oss repos`, `find presentations`, `find public slides`, `find docs and specs`, `find source candidates`, `download candidate`, `source candidate`, `acquisition status`, `자료 후보`, `출처 후보`, `arxiv 링크`, `arxiv 링크 찾아`, `논문 데이터셋 찾아`, `깃허브 저장소 찾아`, `공개 발표자료 찾아`, `문서 스펙 찾아`

Catalog Metadata

Category: research Phase: source-acquisition Quality tier: source-acquisition-gated Reasoning demand: standard

Quality bar:

• Name source kinds from: paper, web_link, dataset, github_repo, presentation, docs_spec, unknown.
• Record acquisition state from: candidate_prepared, link_observed, download_link_prepared, download_observed, file_hash_recorded, text_extraction_observed, license_checked, verification_observed, downstream_selected.
• Separate candidate preparation, observed link, observed download, file hash, text extraction, license check, verification, and downstream selection.
• Attach observation provenance before treating any acquisition state as evidence.
• Vary search angles across official docs, academic work, implementations, datasets, and criticism until each requested source kind has candidates or another angle change adds nothing new.
• Recommend the next downstream workflow without pretending that downstream work already ran.

Required inputs:

• source target or topic
• desired source kinds
• source boundaries or exclusion criteria
• downstream intent when known

Expected outputs:

• source_finder_plan/v1
• source_candidate/v1
• source_candidate_set/v1
• source_acquisition_status/v1
• downstream workflow recommendation
• not-evidence boundary

Artifact expectations:

• source_finder_plan/v1 under .omh/source-finder when a wrapper or CLI records it

Safety rules:

• Do not claim web search, download, repository clone, file extraction, file hash verification, license verification, or source correctness from a prepared candidate.
• Do not redefine research-department's source_inbox/v1; source-finder owns source_candidate_set/v1 and source_acquisition_status/v1 only.
• Route current citations and source-backed synthesis to research, supplied-paper explanation to paper-learning, recurring monitoring to research-department, file export to materials-package, and image cards to img-summary.

Runtime Evidence

Use the current host's own tools and subagent/task mechanism when available; otherwise run the same lanes sequentially or name the unavailable capability. A prepared plan, handoff, checklist, or skill installation is not execution, review, CI, merge-readiness, or merge evidence. Record actual tool results, or not_observed / not_available, in the record; never invent dispatch or host accounting. Treat supplied context as advisory, not proof of hidden memory reads or writes. State scope, constraints, verification, and the stop condition before work. Reply in the user's own words and the host's own voice: OMH's record terms (surface, lane, wrapper, handoff, evidence boundary, not_observed) stay in records and tool calls, never in the sentence the user reads unless they ask about one; and when a stop condition or a decision the user owns ends the turn, offer the next action as a question rather than declaring what will not be done. Supporting paths are relative to this skill directory; sibling skill paths are relative to its parent. Resolve them from the host-provided skill base directory ({baseDir} on hosts that provide it), never a hardcoded install location. A named workflow not installed here is unavailable, not permission to emulate its host-specific capabilities. Verify through the real surface before done.