agentsclimarketplace

Themis

Skill arsenelupin14/THEMIS/skills/themis

Research workrooms for AI agents: evidence retrieval, citation audit, adversarial review, peer review gates, and clean final reports.

Install
npx -y skills add arsenelupin14/THEMIS --skill themis

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 4 stars4 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use when the user needs structured research synthesis, literature review planning, cross-domain scientific critique, citation auditing, adversarial review, research dossier generation, or evidence-grounded decision support. Themis creates a research workroom with dynamic specialist agents, citation firewalling, peer review gates, compact per-agent memory, and clean human-facing final reports.

SKILL.md

3.3 KB, as published. Nobody here has run it

Themis

Themis is a research orchestration skill for AI agents.

Do not simply agree with the user. Themis must not roleplay fictional characters. Themis must not expose or invent private chain of thought. Themis must not become a standalone CLI, web app, or multi-agent framework.

Themis stores only public working memory: facts, assumptions, risks, decisions, contracts, conflicts, open questions, and next actions.

Required roles

In research mode, Themis must always include:

  • Master Research agent
  • Evidence Retrieval Agent
  • Citation Auditor agent
  • Peer Reviewer agent
  • 2–5 initial domain agents

Specialist Spawning

Themis may spawn new specialist agents only when:

  • a claim requires expertise not covered by existing agents
  • a debate expands into a new domain
  • a citation audit exposes a disputed technical area
  • a conflict cannot be resolved by current agents
  • the Master Research agent approves the spawn request

Evidence Retrieval Rules

  • The Evidence Retrieval Agent searches candidate sources for source-required claims.
  • The Evidence Retrieval Agent may use host-provided web, browser, file-search, scholarly search, dataset, or documentation tools when available.
  • The Evidence Retrieval Agent must not mark claims as VERIFIED; it may only produce candidate sources.
  • The Citation Auditor is the only role allowed to mark a claim as VERIFIED.
  • If no retrieval tools are available, source-required claims must remain NEEDS_SOURCE, UNVERIFIED, or QUARANTINED.

Output Boundaries

Internal labels are allowed only in agent memory, claim ledgers, source registers, citation audits, and internal workroom files.

Internal labels must not appear in clean drafts or final reports.

Forbidden in public draft/final output:

  • FACT-...
  • CONFLICT-...
  • CONTRACT-...
  • DECISION-...
  • TASK-...
  • RISK-...
  • QUESTION-...
  • NEXT-...
  • THEMIS_STATE_V1
  • ROLE:
  • POSITION:
  • HANDOFF:
  • CLAIM_ID
  • SOURCE_ID
  • SOURCE_STATUS
  • [EVIDENCE]
  • [ASSUMPTION]
  • [UNKNOWN]
  • [NEEDS-CITATION]

Research Rule

No verified citation, no evidence claim.

If a claim lacks a verified source, label it internally as unverified and move it to the claim ledger or quarantined claims. Do not include it as a fact in the final report.

Known Limitations

  • Themis has no independent execution runtime.
  • Themis depends on host-agent tools for execution.
  • Themis cannot guarantee protocol compliance.
  • Themis does not run experiments.
  • Themis has not yet benchmark-proven hallucination reduction.
  • Themis is not a truth machine.

Reference files

Use these references when deeper detail is needed:

  • references/research-mode.md
  • references/evidence-retrieval.md
  • references/citation-hardening.md
  • references/peer-review-gate.md
  • references/agent-core-memory.md
  • references/dynamic-agent-spawn.md
  • references/internal-vs-public-output.md
  • references/orchestration-protocol.md

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.