agentleFS
Sign inStart free
lulufoolulufoo/lulu-workflow/eval/SKILL.md

eval-library

Skill3 starsChanged 6 days agoMITView on GitHub

Shared Eval orchestration for lulu-workflow. Coordinates dimension probes and remediation.

The file

We split this skill into the rules it gives an agent

9 rules. About 0 are common, 3 are distinctive, and 6 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 6

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Run $EVAL_CONTROL begin-eval-round.

    line 31
  2. Distinctive:

    If the result reports abandonment, run Abandon Handler and stop.

    line 35
  3. Very specific:

    Do not run check-dimension for a skipped dimension.

    line 37
  4. Very specific:

    Run $EVAL_CONTROL begin-dimension --dim {dim} and pin the result.

    line 43
  5. Very specific:

    If the result reports skip: true, record skip_reason and return.

    line 44
  6. Very specific:

    Run $EVAL_CONTROL begin-dimension-remediation --dim {dim} and pin OPERATION_TOKEN from its dispatch_input.

    line 58
  7. Very specific:

    If the caller’s adapter-config / handoff policy_context.completion_mode is return_to_caller (e.g.

    line 74
  8. Distinctive:

    Stop. Do not dispatch remaining dimensions.

    line 83
  9. Distinctive:

    Leave the L in Evaluating.

    line 84

≈ “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

eval/SKILL.md

Coordinates dimension probes and remediation for one round. Done when a probe-only or full-remediation path returns its result.

Script Macros

MacroCommand
$EVAL_CONTROLpython3 "$SKILL_ROOT/eval/scripts/eval_entry.py" --adapter-config-file "$EVAL_ADAPTER_CONFIG" --cycle-id "$CYCLE_ID"

Pin $EVAL_ADAPTER_CONFIG and $CYCLE_ID before $EVAL_CONTROL.

Principles

  • Findings require citable evidence. If evidence cannot be located, escalate; do not guess.
  • A defective SoT is escalated, never silently repaired. A valid SoT that the artifact fails to reflect is remediated.
  • Any non-zero control result is Blocking: stop, report it, and wait for user direction.
  • Remediation processes one dimension at a time.

Begin Eval

  1. Run $EVAL_CONTROL begin-eval-round. Pin the result.
  2. For each returned dimension, run Single dimension (launch) and pin any probe handle.
  3. Await every pinned probe handle. Do not check a dimension before every handle completes.
  4. For each dimension that launched a probe, run $EVAL_CONTROL check-dimension --dim {dim}.
    • If the result reports abandonment, run Abandon Handler and stop.
    • On a non-zero result, apply Blocking and stop.
    • Do not run check-dimension for a skipped dimension.
  5. A probe-only caller runs $EVAL_CONTROL complete-probe-only and stops.
  6. A full-round caller continues to Remediation.

Single dimension (launch)

  1. Run $EVAL_CONTROL begin-dimension --dim {dim} and pin the result.
  2. If the result reports skip: true, record skip_reason and return. Do not pin an operation context, do not dispatch a probe runner, and do not later check this dimension.
  3. Pin the returned operation context.

  4. Dispatch dimension-probe-runner asynchronously with the pinned operation context and this instruction:

    Load dimension-probe-runner/SKILL.md and follow its instructions.
  5. Pin the returned handle and return immediately to Begin Eval.

Remediation

  1. Run $EVAL_CONTROL begin-remediation and pin the result.
  2. Unless the result skips remediation, process each returned dimension serially:
    1. Run $EVAL_CONTROL begin-dimension-remediation --dim {dim} and pin OPERATION_TOKEN from its dispatch_input.

    2. Dispatch remediation-runner synchronously with that pinned context and this instruction:

      Load remediation-runner/SKILL.md and follow its instructions.
    3. Run $EVAL_CONTROL check-dimension-remediation --dim {dim}.
    4. If the result reports abandonment, run Abandon Handler and stop.
  3. Each dimension submits one complete application. Partial issue coverage is Blocking.
  4. Run $EVAL_CONTROL remediation-complete.
  5. Continue to Completion.

Completion

  1. Present the remediation-complete result.
  2. If the caller’s adapter-config / handoff policy_context.completion_mode is return_to_caller (e.g. Compose Atomize Eval), stop here and return the remediation-complete payload to the caller. Do not present Accept L / Fix L / Re-evaluate / Deliver package.
  3. Otherwise ask the user to choose Accept L, Fix L, or Re-evaluate; then exit Eval and return that choice. The caller owns the L transition.

Abandon Handler

  1. Stop. Do not dispatch remaining dimensions.
  2. Leave the L in Evaluating. Report abandonment and wait for user direction.

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.

Check your setup, free