agentleFS
niksavisniksavis/handily/.claude/skills/hill-climb/SKILL.md

hill-climb

Skill2 starsChanged 35 days agoMITView on GitHub

Improves a plan, design, code or skill in graded rounds: fresh reviewers grade fixed properties 0 to 10, the author revises, and a script decides when to stop. Use when asked to hill-climb, iterate to a better version or improve something in rounds.

11 other projects use a version of this, 1 of them edited.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 11 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

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

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

    Record: Put the round's replies in a new folder DIR/rN and run:

    line 22
  2. Distinctive:

    Follow measurements for the manifest and its limits.

    line 30
  3. Distinctive:

    Ask the user only after the climb, unless a finding needs a fact only the user has.

    line 46

“Common” is measured: kept by at least 6 in 10 of the 11 edited versions of this file we could read. “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
<!-- Generated by `basicly skills-build` from skill.yaml. Do not edit; edit the source. -->

Hill Climb

Improve one artifact in graded rounds until the grades stop rising. This skill uses the swarm steps; read that skill first.

The script is ${CLAUDE_SKILL_DIR}/../swarm/scripts/swarm.py, in the swarm skill folder beside this one. DIR is a new folder per climb in the session scratchpad.

Steps

  1. Round 0. Choose 3 to 9 properties from property sets, a subset or your own, named in lowercase with hyphens. Write one line per property that says what a 10 means. Start the state:

    python3 -B ${CLAUDE_SKILL_DIR}/../swarm/scripts/swarm.py climb-init --state DIR/climb.json --properties coverage,done-states,poka-yoke
  2. Grade. Save the version as DIR/vN, where N is the graded round, starting at 1, and run the swarm steps Brief, Launch, Collect and Verify on it with 3 to 6 lenses, at least one not used in the last round. Each brief's rubric holds the line per property that says what a 10 means; each reply holds lens, grades for every property and findings. climb-round replaces the swarm Check.
  3. Record. Put the round's replies in a new folder DIR/rN and run:

    python3 -B ${CLAUDE_SKILL_DIR}/../swarm/scripts/swarm.py climb-round --state DIR/climb.json --lenses a,b,c --dir DIR/rN --refuted 0

    Set --refuted to the number of important findings Verify refuted. Supply host usage with --usage DIR/rN/usage.json when available. Follow measurements for the manifest and its limits. Missing usage stays unknown. The script prints the decision, grades, finding counts and reported usage. Rounds lists the existing stop rules and refusals.

  4. Revise or stop. On continue, revise the artifact from the findings, note what changed, and go to step 2. On stop, report.
  5. Report. Give the final version, changes, grades, finding counts, reported usage and open decisions. Name which verified finding or unresolved decision each revision addressed. Give one recommendation for each open decision. Report missing measurements and cost estimates explicitly.

Rules

  • Grades and citation counts do not prove usefulness or consumer readiness. Verify findings through the existing swarm steps.
  • The author never grades its own version. Grades come only from reviewer replies.
  • The properties stay the same after round 0. For a new property, start a new climb.
  • The script owns the stop rule. Run past a stop only when the user asks, with --past-stop on each extra round, and say so in the report.
  • Ask the user only after the climb, unless a finding needs a fact only the user has. The user's request for a climb approves its reviewer and Verify agents; run them in batches of at most 8.

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.