| name | paper-search-cli |
| description | Use the local paper-search-cli for academic/patent search, optional generic External Search v1 web_search, source presets and search plans, identifier lookup, artifact acquisition, extraction, material ingest/status, workspace storage/export, provider and registry-subscription management, config/migration/status checks, and MCP serving. Trigger for paper-search-cli, paper-search MCP, mcp serve, provider registries, academic_search, patent_search, patent_detail, search-plan, web_search, resource_lookup, resource_add, resource_pdf, artifact_download, extract, material_ingest, workspace_export, platform_status, batch, doctor, config, migrate, registries, providers, or material-providers. Use the Zotero skill instead for Zotero writes or Zotero Resource Search; do not use this skill for general web search that does not need Paper Search. |
Paper Search CLI
Use this skill for the local paper-search-cli system and its MCP server. Run the skill-local launcher for deterministic terminal evidence; use mcp serve only when the request needs the JSON-RPC transport over the same core behavior.
Installed Entrypoint
Run commands from this skill directory through its launcher:
node scripts/paper-search.mjs --version
node scripts/paper-search.mjs paths --json
node scripts/paper-search.mjs self status --json
node scripts/paper-search.mjs self mode --json
node scripts/paper-search.mjs self update --json
The installed skill is a Junction or symlink into a retained Paper Search
checkout. This launcher resolves that checkout and starts its verified runtime.
Do not bypass it with a repository-relative node dist/cli.js command during
normal agent use. The paper-search command is the equivalent shim for humans.
Activation Boundary
Use this skill when the request is explicitly about paper-search-cli, its provider packages, its local workspace, its material artifact/extraction records, its MCP server, or its source-compatible canonical tools.
Do not use this skill for ordinary news lookup, broad web browsing, or general current-events search unless the user explicitly asks to route that work through paper-search-cli.
Eight Capability Groups
| Group | Use for |
|---|
discover | Academic and patent search, plus optional generic External Search v1. |
identify | DOI/PMID/arXiv/ISBN/URL metadata lookup and patent detail. |
assess | Reserved group with no implemented tools; do not route work here (promotion criteria: ADR-0003). |
acquire | Artifact acquisition/recording from a URL, workspace item, or DOI, plus PDF attachment records. |
extract | Markdown/JSON/assets extraction from an artifact, URL, or local file through material providers. |
organize | Local workspace add/list/export operations. |
orchestrate | Batch rows, material ingest, and material status over stored records. |
operate | Status, search-plan, config, provider management, help/tools, and MCP serving. |
Reference Rule
Before parameter-heavy work, read the matching reference:
- Read
references/capability-routing.md before mapping a request to the eight groups and current entrypoints.
- Read
references/cli-contract.md before touching working directory, build/probe, JSON output, ResultEnvelope, run <canonical_tool>, aliases, batch rows, dry-run/plan, local records, or secrets/config.
- Read
references/management-layer.md before using status, doctor, search-plan, source presets/config fragments, config, migrate, registries, subscription-bound providers available|install|update, providers inventory, providers --kind search|material, material-providers, mcp serve, platform-status, help, tools, or smoke gates.
- Read
references/management-layer.md before planning or applying self mode
or self update. These commands are plan-first. Production authority is
source-sealed to the official HTTPS main origin and cannot be changed by
config, environment, or CLI input; other clones remain user-managed.
For material download, extraction, or ingest, route through material-provider discovery and a dry-run/plan before live network or write execution. DOI inputs are resolved to candidate URLs through installed resolver providers before download. No networked material resolver, downloader, or extractor is built into the core CLI; installed material providers must be discovered and validated.
Trust Live CLI
If this skill or a reference disagrees with the installed CLI, trust the live
CLI and report the documentation drift. Run every probe below from this skill
directory. If the launcher reports a missing or incompatible verified build,
follow its installer command rather than invoking dist/cli.js directly.
Prefer these probes:
node scripts/paper-search.mjs --version
node scripts/paper-search.mjs tools --json
node scripts/paper-search.mjs help
node scripts/paper-search.mjs status --json
Validation Boundary
Report the exact command evidence used. Keep smoke or live-network validation explicit: do not claim live provider, external web search, remote PDF, networked material extraction, or MCP behavior unless the matching smoke gate, provider inspection, contract test, or JSON-RPC call was actually run. status is static; doctor may run only the External Search v1 no-network probe.