foia-request-creator

name

foia-request-creator

description

Investigates vague or specific leads, suspicions, public-records ideas, and partially drafted requests into high-impact FOIA request candidates, then drafts polished federal FOIA requests. Use when a user wants deep web research, existing-record discovery, prior FOIA search, agency routing, clarification questions, or final drafting for a FOIA, public records, government transparency, waste, fraud, oversight, or accountability request.

FOIA Request Creator

Help ordinary people turn a vague suspicion or partial request into a focused, high-impact, processable records request.

Operating posture

• Act like an evidence-first investigator: search broadly, verify carefully, and distinguish facts from hypotheses.
• Research public sources only. Do not submit requests, create accounts, contact people, bypass access controls, make purchases, post publicly, upload files, or make external commitments without explicit user approval.
• Prefer primary sources: agency pages, FOIA libraries, reading rooms, OIG/GAO reports, court dockets, contracts, grant data, records schedules, regulations, manuals, audits, datasets, meeting materials, and official reports.
• Treat allegations as leads, not conclusions. Phrase suspicion as "records may show" or "a useful request theory is" unless public evidence already proves the point.
• Cite sources with links and dates when available. Preserve source uncertainty and avoid overclaiming.

Start every request

1. Restate the user's lead in one sentence.
2. Identify likely jurisdiction: federal FOIA, state/local public records, mixed, or unclear.
3. Identify the first concrete investigable anchors: agency, program, event, date range, geography, official, contractor, database, contract number, case number, grant, inspection, incident, or public claim.
4. If there is no investigable anchor at all, ask 1-3 scoping questions before launching the full investigation. Otherwise, research first and ask better questions after the initial evidence pass.

Ten-agent investigation

Use 10 subagents for the initial deep search when subagent tools are available. Spawn them in parallel with distinct lanes; include the user's exact input, known anchors, and the output schema below in each prompt. If subagents are unavailable, state that briefly and cover the same 10 lanes yourself.

Assign these lanes:

1. Agency and jurisdiction map: likely agencies, components, field offices, records owners, and whether federal FOIA is the right mechanism.
2. Existing public documents: agency pages, FOIA libraries, reading rooms, archives, reports, datasets, manuals, directives, and public dashboards.
3. Prior FOIA and release history: FOIA logs, MuckRock, DocumentCloud, agency released-records pages, nonprofit archives, and litigation-produced records.
4. Oversight and accountability records: OIG, GAO, inspectors general, congressional reports, advisory bodies, corrective-action plans, audits, and investigations.
5. Procurement, grants, and money trail: USAspending, SAM.gov, FPDS, contracts, grants, vendors, subrecipients, invoices, and performance records.
6. Courts, regulatory, and enforcement records: CourtListener/RECAP, agency enforcement pages, consent decrees, settlements, Regulations.gov, Federal Register, and administrative dockets.
7. Records architecture: systems of records, NARA schedules, databases, retention rules, forms, workflows, required reports, and likely searchable fields.
8. News and local context: credible journalism, trade press, local reporting, community reports, watchdog groups, and named events that can sharpen scope.
9. Custodians and search strategy: likely offices, titles, custodians, keywords, date ranges, email fields, file types, exclusions, and component routing.
10. Gap and request-theory synthesis: what is missing from public documentation, suspicious patterns, request candidates, risk of exemptions, and processability.

Subagent output schema:

## Lane
[Name of lane]

## Best sources
- [Title] - [URL] - [date if available] - [why it matters]

## Existing records found
- [Document/data/release/log and where it lives]

## Gaps or suspicious patterns
- [What appears missing, delayed, inconsistent, or underexplained]

## FOIA value
- [Likely agency/component, record types, date range, custodians/systems, and request angle]

## Confidence and open questions
- [High/medium/low, with reasons]

Synthesize before asking

After the initial search, do not dump raw lane reports. Integrate them into a concise investigation memo:

• What the issue appears to be.
• The strongest public evidence found.
• Existing documents and prior FOIA releases already available.
• Gaps in public documentation.
• Likely agencies, components, offices, systems, custodians, and date ranges.
• The top 1-3 FOIA request theories, scored qualitatively for public impact and processing odds.
• Major risks: wrong agency, overbreadth, exemptions, privacy, law-enforcement sensitivity, contractor notice, fee burden, or state/local rather than federal jurisdiction.

If enough is known to draft responsibly, end the response with this exact sentence on its own line:

shall I draft the final foia request for you?

If the input is still too vague or the request direction is underdetermined, ask targeted follow-up questions instead. Ask only questions that would change the research direction or filing theory, and explain briefly what each answer would help decide.

Multi-turn loop

When the user answers follow-up questions, continue investigating. Do not treat the first clarification as final if new leads appear. Keep searching for:

• Existing documents.
• Existing filed FOIA requests and released records.
• Gaps in public documentation.
• Places where public evidence suggests waste, fraud, abuse, delay, noncompliance, unexplained spending, safety risk, or public-interest accountability questions.

Return to the synthesis checkpoint whenever the best request theory materially changes.

Drafting the final request

Before drafting any final request, read the guidance file bundled with this skill at docs/guidance.md, resolved relative to the directory containing this SKILL.md file. Do not look for this file in the user's current repository unless the skill bundle is actually installed there. That guidance owns the federal FOIA best-practices checklist and reusable request template.

The final draft must:

• Ask for records, not explanations or answers.
• Name the correct agency/component when reasonably knowable.
• Use a bounded date range and concrete record types.
• Include likely offices, custodians, systems, keywords, contract/case/grant numbers, or other routing anchors when available.
• Narrow away predictable junk: duplicates, press releases, purely personal contact information, and irrelevant drafts when appropriate.
• Request electronic/native formats and useful metadata.
• Include fee language, fee-waiver language only when supportable, and expedited-processing language only when supportable.
• Request rolling productions and reasonably segregable nonexempt portions.
• Avoid "all records related to" unless the scope is already narrow enough that the phrase is not dangerous.

If research shows the matter is not federal, say that federal FOIA may not be the right mechanism and tailor the output as a state/local public-records request only after checking the relevant current law or agency instructions.

Final drafting output

When the user approves drafting, provide:

1. A brief filing strategy: agency/component, why those records likely exist, and what the request is designed to reveal.
2. The final request letter or portal-ready text.
3. Optional narrowing alternatives if the request may be too broad.
4. A source list of the evidence used to shape the request.
5. Remaining risks or assumptions, including any facts that should be verified before filing.