agentleFS
markoblogomarkoblogo/abvx-agent-skills/skills/evidence-ledger-research/SKILL.md

evidence-ledger-research

Skill16 starsChanged 5 months agoMITView on GitHub

Research and answer evidence-sensitive questions with source, date, unit, and operand discipline. Use for web research, PDF/document QA, policy or finance facts, tables, time series, calculations from retrieved evidence, exact-answer extraction, and any task where wrong source selection, stale facts, nearby table confusion, or formatting errors would matter.

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 8 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈8

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill when the answer must be tied to exact evidence rather than memory or broad summaries.

    line 12
  2. Distinctive:

    Prefer primary sources, official data pages, original PDFs, repository files, or provided/oracle documents.

    line 16
  3. Distinctive:

    If a guessed exact-date search fails, broaden to the official series, table name, dataset code, or download page.

    line 17
  4. Distinctive:

    Read only the narrow span needed after locating a promising source, then verify entity, period, basis, unit, and version.

    line 18
  5. Distinctive:

    If a provided document contains the answer, extract from it before searching elsewhere.

    line 19
  6. Very specific:

    If evaluating Doc-to-LoRA or another document-memory method, pair this with doc-to-lora-evaluator and keep the baseline questions, source …

    line 21
  7. Distinctive:

    Use the ledger before arithmetic, comparisons, ranking, or final formatting.

    line 37
  8. Distinctive:

    Do not round intermediate values unless the source requires it.

    line 53
  9. Distinctive:

    When citing sources, link the source and summarize the evidence without long quotations.

    line 59

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Evidence Ledger Research

Use this skill when the answer must be tied to exact evidence rather than memory or broad summaries.

Source Discipline

  • Prefer primary sources, official data pages, original PDFs, repository files, or provided/oracle documents.
  • If a guessed exact-date search fails, broaden to the official series, table name, dataset code, or download page.
  • Read only the narrow span needed after locating a promising source, then verify entity, period, basis, unit, and version.
  • If a provided document contains the answer, extract from it before searching elsewhere.
  • For volatile facts, verify the latest source before answering.
  • If evaluating Doc-to-LoRA or another document-memory method, pair this with doc-to-lora-evaluator and keep the baseline questions, source answers, and negative controls in the ledger.

Evidence Ledger

Maintain a compact ledger for multi-step answers:

Claim/operand:
Source:
Date/period:
Unit/scale:
Role:
Value/span:
Notes:

Use the ledger before arithmetic, comparisons, ranking, or final formatting. Do not compute from partial evidence unless the user explicitly accepts an estimate.

Table And Document Rules

  • Match row labels and exact column headers, not visual proximity alone.
  • Watch for repeated month rows, fiscal vs calendar sections, continuation tables, estimates, totals, and footnotes.
  • For forms and PDFs, locate the exact field/anchor in the question, then read the adjacent filled value from the same row, box, block, or table cell.
  • For visual/document QA, prefer the smallest exact text span that answers the question.
  • Preserve spelling, punctuation, capitalization, and numeric formatting when exact extraction matters.

Calculation Rules

  • Confirm every operand before calculating.
  • Track scale and direction: thousands/millions/billions, currency, exchange-rate orientation, sign, percent vs percentage points.
  • Preserve relation direction: "change from A to B" means B - A; "share accounted for by X" means X / total.
  • For time windows, enumerate expected periods first and verify endpoint inclusion.
  • Do not round intermediate values unless the source requires it.

Final Answer

  • Match the requested format exactly: option label, numeric value, list, date, or concise phrase.
  • Include unit words only when requested or necessary for clarity.
  • When citing sources, link the source and summarize the evidence without long quotations.
  • Copy the final answer from the checked ledger value, not from an unverified intermediate.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.