agentleFS
selamy-labsselamy-labs/agent-skills/skills/evidence-claim-boundaries/SKILL.md

evidence-claim-boundaries

Skill0 starsChanged 7 days agoMITView on GitHub

Use when verifier output, schemas, fixtures, templates, CI, or readiness artifacts are being used to support live, human-gated, hardware, deployment, or completion claims.

The file

We split this skill into the rules it gives an agent

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈9

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill when a task produces or reviews evidence artifacts that could be mistaken for proof of the real-world outcome.

    line 8
  2. Distinctive:

    If evidence is readiness or proxy, say what it prepares or checks.

    line 25
  3. Distinctive:

    When reporting progress, include:

    line 30
  4. Distinctive:

    Prefer this compact form:

    line 37
  5. Distinctive:

    Keep a completed execution's frozen package and manifest inputs unchanged.

    line 43
  6. Distinctive:

    If later preparation changes a recorded input, preserve both versions and the accepted result.

    line 47
  7. Distinctive:

    When the human-gated artifact exists, report the required bundle:

    line 68
  8. Distinctive:

    Before marking an issue, PR, or task complete, ask:

    line 77
  9. Distinctive:

    If the required artifact is unavailable, is the task reported as waiting, blocked, or ready for human attempt rather than complete?

    line 84

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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 Claim Boundaries

Use this skill when a task produces or reviews evidence artifacts that could be mistaken for proof of the real-world outcome. Verifiers, schemas, fixtures, templates, summaries, and CI runs can make a future check stronger, but they do not prove that the underlying live run, user-visible behavior, hardware path, or deployment outcome happened.

Classify the Evidence

Name the evidence category before writing the claim:

  • readiness: tooling, verifier, schema, template, fixture, or docs hardening.
  • proxy: logs, summaries, screenshots, traces, or generated reports that describe another artifact.
  • primary: the real artifact or behavior in the environment where it matters.
  • human-gated: a primary artifact that requires a person to perform or observe the live run.

If evidence is readiness or proxy, say what it prepares or checks. Keep the completion claim open until primary evidence exists.

Write Bounded Claims

When reporting progress, include:

  • The artifact changed or checked.
  • The verifier or review path used.
  • The exact outcome that is supported.
  • The outcome that is not supported yet.

Prefer this compact form:

Readiness landed: <artifact/check>. Supports: <future claim it can verify>. Not claimed: <live/human/hardware/deployment outcome still missing>.

Preserve Completed Packages

Keep a completed execution's frozen package and manifest inputs unchanged. Prepare follow-on edits and outputs in a separate package outside the recorded scope, including any directory or prefix entries; reuse immutable inputs.

If later preparation changes a recorded input, preserve both versions and the accepted result. When existing evidence establishes that the accepted execution used verified original bytes and the drift happened afterward, restore those bytes and verify their recorded identity. Record the drift and restoration identities and observed timing; restoring bytes must not erase the intervening event. Retain that acceptance instead of replaying accepted effects to repair file drift. If the originals cannot be verified, the executed inputs differed, or the effects are uncertain, use the existing recovery or affected requalification path.

Carry out routine reversible artifact repairs covered by standing authority without repeated per-file escalation. Preserve explicit holds, source custody, and human gates; changed executed artifacts still need their affected checks.

Human-Gated Outcomes

For manual or human-gated acceptance, stop at "ready for human attempt" until the person-produced sanitized artifact exists. CI, fixtures, dry runs, templates, or simulated inputs can prepare the path, but they cannot stand in for the manual run.

When the human-gated artifact exists, report the required bundle:

  • Same-session or same-environment identifier, if the claim depends on it.
  • Sanitized primary artifact location or attachment.
  • Verifier result or reviewer-safe summary.
  • Residual missing rows, if any.

Review Checklist

Before marking an issue, PR, or task complete, ask:

  • Is this primary evidence, or only readiness/proxy evidence?
  • Does the wording name the real artifact still required?
  • Could a reader mistake this for live, hardware, deployment, or human acceptance?
  • Are synthetic, fixture, dry-run, or CI-only artifacts clearly labeled as such?
  • If the required artifact is unavailable, is the task reported as waiting, blocked, or ready for human attempt rather than complete?

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.