Company Research

Purpose

Use this skill to start and monitor Primr research runs through MCP. Treat MCP as the source of truth for current modes, defaults, and outputs.

Workflow

1. Read primr://research/modes to confirm the current mode guidance.

1. Read primr://research/status to see whether a run is already active.

1. Call estimate_run before starting a new run. Supports platforms (array), strategy_type, and no_ai_strategy parameters. Returns estimated_time_range alongside estimated_time_minutes.

1. Present the estimate, state that the run incurs real API cost, and get explicit approval before calling research_company.

1. After starting a run, monitor with wait_for_status_change or check_jobs until terminal.

1. Read primr://output/artifacts/by_job/{job_id} to inventory completed-job

artifacts without report body content, then read primr://output/latest only when a report preview is needed to present the result and next actions. Read primr://output/qa_summary/by_job/{job_id} when QA artifacts are attached and the user needs QA outcome metadata. Read primr://output/usage_summary/by_job/{job_id} when the user needs cost, timing, approval, or artifact-count metadata. Read primr://output/source_summary/by_job/{job_id} when the user needs citation/source appendix metadata. Read primr://output/trace_summary/by_job/{job_id} when the user needs scrape trace health metadata. Read primr://output/verification_summary/by_job/{job_id} when the user needs claim verification trust score, claim counts, status counts, first-party downgrade counts, or source-reference counts without raw claims, source URLs, search queries, explanations, or report body content. Read primr://output/calibration_summary/by_job/{job_id} when the user needs label-calibration counts, inference source-copy counts, evidence-review count buckets, judge provenance, or judge-agreement metadata without raw claims, source URLs, evidence reviews, rationales, or report body content. MCP resource reads are audit-logged with hashed URI/result values and normalized resource kind, without raw URI query values or resource bodies. A2A skill calls are audit-logged with hashed message/result payloads, hashed caller ids, granted scopes, duration, outcome, and job id when present, without raw message text, task ids, URLs, report paths, raw results, or caller ids.

Operating Rules

• Prefer the lightest mode that answers the request.

• Do not restate fixed pricing or provider defaults from memory. Use MCP resources and estimate_run.

• If a job is already in progress, surface that state instead of starting a second run.

• Treat full as the standard end-to-end workflow and premium as the higher-cost, higher-depth option.

• Expect standard runs to take roughly 35-50 minutes and premium multi-vendor runs to reach 75-120 minutes.

• Do not hold the client in a synchronous wait for the entire run; launch once, then monitor and resume from job state.

Example

User: Research ExampleCo at https://example.com



1. Read primr://research/modes

2. Read primr://research/status

3. estimate_run(company_url="https://example.com", mode="full")

4. After approval: research_company(company_name="ExampleCo", company_url="https://example.com", mode="full", max_estimated_cost_usd=0.67)

5. wait_for_status_change(job_id="...", timeout_seconds=60)

Error Handling

• job_in_progress: report the active run and offer to monitor it.

• invalid_url or ssrf_blocked: ask for a valid public URL or switch to deep if appropriate.

• rate_limit_exceeded: wait for the retry window and retry.