Skip to content
Agentmart
Submit a tool

research

Answer focused questions from documents, APIs, or a few sources; route broad comparisons and high-stakes investigations to deep-research.

Skill by lstmemery

IndexedMIT★ 1
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/lstmemery-research

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Answer focused questions from documents, APIs, or a few sources; route broad comparisons and high-stakes investigations to deep-research.

Use cases: research

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/research

What's inside

Skill name
research
Description
Answer focused questions from documents, APIs, or a few sources; route broad comparisons and high-stakes investigations to deep-research.
License
MIT
Instructions
77 lines in SKILL.md · view on GitHub

Files (5)

  • EVIDENCE.md
  • REFERENCE.md
  • SKILL.md
  • openai.yaml · agents/
  • trigger-cases.json · tests/

SKILL.md (first lines)

# Research

Answer the user's focused question from opened sources. Finish when each material
claim matches its evidence, contradictions and missing evidence are explicit,
and the active runtime's output and delivery contract is satisfied.

Load the active runtime's durable user preferences and delivery contract for
this step. When this run produces a finished report, deliver it through that
contract by default after the research coordinator has completed synthesis and
validation. Do not ask again whether to send it. Honor an explicit opt-out in
the current request. Workers return their slice artifacts to the coordinator
and never deliver them. Report only the delivery state observed; say the report
was emailed only after the delivery helper verifies receipt.
For jailed research's independent fact-check requirement before PrivateBin/PDF
delivery, follow the [delivery rule](REFERENCE.md#jailed-research-delivery).

Use the request first, then relevant supplied prior art and sources that own
the claims: official documentation, code, specifications, APIs, or original
papers. Search snippets and model memory are leads. A prior report is a
`[PRIOR-NOTE]`, not fresh verification; reopen material or changeable claims.

## Choose the path

- **Single lookup:** one fact or owner document. Inspect relevant supplied
  context, open the source, answer from the supporting passage, cite it, and
  state material limitations. Compare the final wording, units, qualifiers,
  and date with that passage before returning. Keep the evidence with the
  answer; no separate plan, evidence table, or audit artifact is required.
- **Focused investigation:** several disputed claims, a few sources, or a short
  causal chain. Follow the investigation below.
- **Broad or high-stakes investigation:** load and follow skill `deep-research`
  before gathering sources. Scope, rather than a fixed source count, decides.

A required saved report may be the same concise answer as a single lookup.
Use the runtime's output root and delivery mechanism. Queue only the finished
report, never plans, drafts, or worker slices. If no delivery mechanism exists,
say so and leave the report available. Confirm only the delivery state
observed.

## Focused investigation

1. **Scope.** Inspect supplied prior art before searching. Record the question,
   audience, recency boundary, relevant definitions, exclusions, and stop rule
   in a compact note. Infer ordinary details; ask only when a missing decision
   would materially change the answer.
2. **Gather.** Read [EVIDENCE.md](EVIDENCE.md) for the shared source, evidence-row,
   and claim-check contract. Find and open owner sources. Keep one evidence
   table with support or an explicit gap for each requested part.
3. **Synthesize and check.** Map findings to evidence; resolve contradictions
   or carry both claims into the gap statement. Separate recommendations from
   source claims. Apply the shared claim check to all retained material claims,
   including uncited prose. A captured passage can satisfy the check; fetch
   again when freshness, incomplete capture, or a contradiction requires it.
4. **Deliver.** Use [REFERENCE.md](REFERENCE.md#investigation-report) for the
   report fields. Stop when the scoped parts are supported or unresolved,
   further retrieval repeats existing evidence, and the claim check passes.
   Preserve limitations rather than implying that missing evidence proves a
   negative. Deliver through the active runtime's contract.

For long, independent reading that earns a worker's overhead, or explicit
user-requested delegation, read [REFERENCE.md](REFERENCE.md#worker-contract).
A single-owner fact stays inline or uses at most one worker. If scope grows,
record why the route changed.

## Continuity and recovery

Across turns, preserve scope, selected route, evidence/artifact locations,
decisions, contradictions, gaps, check results, and the next action. Keep these
in the existing run artifact rather than creating a second state system.

For retrieval failure, try one transient retry or an alternate owner source;
record a continuing failure as a gap. Read
[REFERENCE.md](REFERENCE.md#retrieval-and-worker-recovery) when retrieval or a
worker result is incomplete. Missing evidence may narrow the answer; it cannot
justify an unsupported claim.

From lstmemery/skills, MIT.

More from lstmemery/skills

Signals

Stars
1
Forks
0
Open issues
0
Last commit (this folder)
Oct 12, 2026 (16 days ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from lstmemery

Cite this page

research skill. Agent Mart. https://agentmart.dev/skills/lstmemery-research (last updated 2026-10-24)