Skip to content
Get packs

HillClimb Skill Improvement Loop

Iteratively improves another agent skill by testing candidate changes against real failures and keeping only measured wins.

Source Download skill
Repository
danielmiessler/LifeOS
Stars
19,682
Licence
MIT
Files
11 files

What it does

This skill makes an AI agent run a disciplined improvement loop on one existing skill. It gathers real failures (user complaints, failing test cases, bad transcripts), turns them into test cases, proposes candidate edits, and keeps a change only when a measured test run shows improvement. Learnings are recorded as dated notes or standing checks rather than longer prompts, and the whole history is committed and tracked. The files also define tools and safety gates, such as not weakening existing checks and running consumer tests before changing a shared output format.

Great for

  • Maintainers of a LifeOS-style personal agent system who want a skill's complaint rate to go down measurably
  • Developers who already have test suites and error logs for an agent skill and want a structured tuning process
  • Teams that want changes to agent skills tracked with before/after statistics rather than guesswork

Skip it if

  • Someone without an existing skill, test suite, or failure log: the loop depends on real misses and a baseline measurement, and the files say a suite must be built first
  • Casual one-off prompt tweaks: the process is heavy (baselines, paired comparisons, ledgers) and is designed for sustained climbs
  • Environments where automated edits to skill files are not reviewed: the loop commits kept diffs and edits skill content, so unreviewed use could change behaviour unexpectedly

Try asking

  • “Start a HillClimb run on my Invoice skill, targeting its test suite pass rate, using the complaints from last month as the reason”
  • “Show me the history of the climb for the Research skill so I can hand it to a proposer”
  • “Check whether the Summary skill's complaint rate moved since the last committed change”
  • “Run preflight on this climb and tell me whether the suite can detect a meaningful win”
Step-by-step guide: how to use HillClimb Skill Improvement LoopInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 11 original files, the MIT licence and an attribution note.

    Intents.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip Intents.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip Intents.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Build the test suite from real misses first and run preflight before round 1, since the files say a climb without a detectable win should add cases before proposing edits
  • Keep the change history and standing checks intact; the rules forbid deleting or weakening checks to pass a climb
  • Review the committed diffs and Gotchas notes yourself, because the loop edits skill files and runs scripts on your machine

What's inside

  • Eval.md
  • Ideate.md
  • IsaClimb.md
  • Loop.md
  • Metric.md
  • Miss.md
  • Prompt.md
  • Skill.md
  • Survey.md
  • Verify.md
  • Workflow.md

Redistributed unmodified from danielmiessler/LifeOS under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.