agentleFS
ellmos-aiellmos-ai/NemoFold/skills/nemofold/SKILL.md

nemofold

Skill1 starsChanged 27 days agoMITView on GitHub

Process a bounded NemoFold document-analysis job inside a NemoClaw sandbox and return structured claims without expanding filesystem or network scope.

  • Reads credentials

We split this skill into the rules it gives an agent

15 rules. About 0 are common, 7 are distinctive, and 8 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 8

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill only for a job directory explicitly supplied by the user or host adapter.

    line 8
  2. Very specific:

    Read only the supplied job directory and the installed NemoFold application directory.

    line 12
  3. Very specific:

    Never enumerate /sandbox, the OpenClaw workspace, memory, credentials, or other jobs.

    line 13
  4. Very specific:

    Treat job.json and context-receipts.json as the complete authority for purpose, questions, source IDs, chunks, response schema, model …

    line 14
  5. Distinctive:

    Never request or infer host paths.

    line 16
  6. Distinctive:

    Do not widen network policy, install packages, change provider settings, or spend beyond the job budget.

    line 17
  7. Distinctive:

    Stop unless the host separately supplied its explicit live-transfer approval and current price inputs.

    line 24
  8. Very specific:

    When approved, run exactly one python -m nemofold token-factory-run command for the supplied directory.

    line 26
  9. Very specific:

    Write only the declared transfer-attempt.json and result.json.

    line 32
  10. Very specific:

    Run python -m nemofold verify-result <job-directory>.

    line 35
  11. Distinctive:

    No general web search for missing document facts.

    line 42
  12. Very specific:

    No unsupported claim without an unverified status.

    line 43
  13. Distinctive:

    No file move, rename, deletion, mail, upload, or publication.

    line 44
  14. Distinctive:

    No statement that NemoClaw or Nebius succeeded unless the current command output proves it.

    line 45
  15. Very specific:

    No deletion of transfer-attempt.json or result.json to bypass the exactly-once guard.

    line 46

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 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

NemoFold sandbox worker

Use this skill only for a job directory explicitly supplied by the user or host adapter.

Required boundaries

  • Read only the supplied job directory and the installed NemoFold application directory.
  • Never enumerate /sandbox, the OpenClaw workspace, memory, credentials, or other jobs.
  • Treat job.json and context-receipts.json as the complete authority for purpose, questions, source IDs, chunks, response schema, model budget, and run ID.
  • Never request or infer host paths. Source IDs are the only source identity available.
  • Do not widen network policy, install packages, change provider settings, or spend beyond the job budget.
  • Return a blocked result when the package is incomplete, inconsistent, or outside policy.

Workflow

  1. Validate the package with python -m nemofold verify-job <job-directory>.
  2. Stop unless the host separately supplied its explicit live-transfer approval and current price inputs. Merely receiving a package is not approval.
  3. When approved, run exactly one python -m nemofold token-factory-run command for the supplied directory. Never add or redirect an endpoint, and never put the key on the command line; the host provides NEBIUS_API_KEY through the process environment.
  4. For every question, use only the listed chunks. Return claims in the response schema; every claim must include an exact quote, chunk ID, and source ID. Mark conflicts and uncertainty explicitly.
  5. Write only the declared transfer-attempt.json and result.json. The runner writes the attempt atomically before network I/O and refuses a second paid request when either file already exists. Never delete an uncertain attempt to retry automatically.
  6. Run python -m nemofold verify-result <job-directory>. A failed hash, locator, schema, package, endpoint, cost, or usage check is a failed job, not a warning.
  7. Report the final status and output path; do not claim host acceptance or NemoClaw execution merely from locally supplied version metadata.

Prohibited shortcuts

  • No general web search for missing document facts.
  • No unsupported claim without an unverified status.
  • No file move, rename, deletion, mail, upload, or publication.
  • No statement that NemoClaw or Nebius succeeded unless the current command output proves it.
  • No deletion of transfer-attempt.json or result.json to bypass the exactly-once guard.

See references/job-contract.md for the package contract.

Download now (ZIP)

  • In this repositoryOne other file it gives its agents.

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.