[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M77M59K0Z14BX4JSJB40BE76":3,"skill-same:01M77M59K0Z14BX4JSJB40BE76":67},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":29,"commitSha":30,"skillDirectory":31,"forksCount":32,"repositoryHtmlUrl":33,"repositoryHomepageUrl":6,"ownerHtmlUrl":34,"ownerBlogUrl":35,"ownerDescription":36,"updatedAt":37,"install":38,"resources":50},"01M77M59K0Z14BX4JSJB40BE76","climb",null,"Keep-or-revert loop on a frozen harness: one hypothesis and one commit per attempt, both arms re-measured every time, kept by a rule written before the first attempt, logged outside the tree, in its own worktree and branch. Use when: 'keep trying changes until this counter drops', 'hill-climb this benchmark', 'iterate on this until the target is met', 'run an optimization loop'. After \u002Fperformance:snapshot baseline, before \u002Fperformance:verify. Skip when: the score comes from an eval set for an LLM app (`\u002Fclaude-api hillclimb`); one before-and-after comparison (`\u002Fperformance:snapshot`).","melodic-software\u002Fclaude-code-plugins","melodic-software","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F58273638?v=4","01M77JVQTXNGKCE4MDFY2BC936","Melodic Software","01M77JVQVN9QDWKAH6W03ANP76",30,"MIT","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fperformance\u002Fskills\u002Fclimb\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\ndescription: \"Keep-or-revert loop on a frozen harness: one hypothesis and one commit per attempt, both arms re-measured every time, kept by a rule written before the first attempt, logged outside the tree, in its own worktree and branch. Use when: 'keep trying changes until this counter drops', 'hill-climb this benchmark', 'iterate on this until the target is met', 'run an optimization loop'. After \u002Fperformance:snapshot baseline, before \u002Fperformance:verify. Skip when: the score comes from an eval set for an LLM app (`\u002Fclaude-api hillclimb`); one before-and-after comparison (`\u002Fperformance:snapshot`).\"\nuser-invocable: true\nargument-hint: \"[\u003Cgoal-slug>]\"\ndisable-model-invocation: false\nmetadata:\n  workflow-stage: implement\n  summary: Keep or revert one measured change at a time until the goal is met\n---\n\n**Arguments.** `[\u003Cgoal-slug>]`. e.g. \u002Fperformance:climb site-build-template-parses\n\n## Purpose\n\nAnswers **\"which of these changes actually moved the number, and how far can it go?\"**\n\nMeasure one change at a time: hypothesis, change, commit, measure both arms, apply the rule, keep\nor revert, log, until the goal is met. Keep every judgment out of reach of the attempt being\njudged: the rule is written first, the harness is frozen, and the baseline is re-measured beside\neach attempt.\n\nRead [`${CLAUDE_PLUGIN_ROOT}\u002Freference\u002Fharness-integrity.md`](${CLAUDE_PLUGIN_ROOT}\u002Freference\u002Fharness-integrity.md)\nbefore the first attempt.\n\n## Inputs, or stop\n\n1. **A goal from `\u002Fperformance:goal`** with its floor, its Realistic target and `min_attempts`. A\n   goal without `min_attempts` goes back to `\u002Fperformance:goal`; never pick one here, since goal is\n   human-gated.\n2. **A harness `\u002Fperformance:snapshot` froze**: the command, the commit it was frozen at, and its\n   discrimination result. Send the run back to `\u002Fperformance:snapshot baseline` when no harness is\n   frozen, or when its output carries neither tally. Each harness run must print three numbers: the\n   goal's counter, a tally of failures and a tally of finished work units.\n3. **A clean checkout.** `git status --porcelain` in your checkout prints nothing. Refuse\n   otherwise and name the files: the loop starts from a commit, and uncommitted work is not in it.\n4. **Optional: a holdout harness `\u002Fperformance:snapshot` froze**, a second command that prints the\n   same three numbers on inputs the main harness does not use. It needs its own `Frozen:` line in\n   the baseline report with `discriminate: exit 0`, worded like item 2; send the run back to\n   `\u002Fperformance:snapshot baseline` when the line is missing. **Absent means off**: the run behaves\n   exactly as it does without this item, and nothing below that mentions the holdout applies.\n\n## Set up, before the first attempt\n\n1. **Memory slice.** Use `\u003Cmemory_dir>\u002F\u003Ctopic-slug>\u002Fclimb\u002F\u003Cgoal-slug>\u002F` in your checkout (default\n   `\u003Cmemory_dir>` is `.work\u002F`), by absolute path: the loop runs in another worktree, and the log\n   must outlive it. Done when the directory exists.\n2. **Keep rule.** Write `keep-rule.json` there with exactly three keys: `counter` (the goal's\n   counter, by name), `direction` (`lower` or `higher`) and `gain_floor` (the change in the counter\n   that counts as a gain; the host's measured noise for a duration, `0` for a deterministic count).\n   Show it to the person when one is present. **Never edit it during the run**: once an attempt\n   has a number, any change to the file would tailor the rule to that number.\n3. **Two worktrees, outside your checkout or in a directory its `.gitignore` ignores.** Start both\n   at your checkout's `HEAD`:\n   - climb's own: `git worktree add -b climb\u002F\u003Cgoal-slug> \u003Cclimb-path> \u003Cstart-sha>`. A branch with\n     that name already exists: stop and ask, since it may hold an earlier run.\n   - the baseline arm: `git worktree add --detach \u003Cbaseline-path> \u003Cstart-sha>`. It always sits at\n     the last kept commit.\n\n   Every attempt happens in `\u003Cclimb-path>`. Nothing in this skill edits, commits in or resets your\n   checkout.\n4. **Starting measurement.** Run the harness once in each worktree. The counters agree, the error\n   counts are zero and the work counts match, or the harness is not stable enough to climb on:\n   stop and say which number differed.\n5. **Holdout, only when item 4 is given.** Write `holdout.json` in the slice, holding the\n   holdout's `Frozen:` command and sha; like `keep-rule.json` it is never edited during the run.\n   It is not a key of `keep-rule.json`, which `decide` accepts only with its three keys. Then run\n   the holdout once in each worktree, by absolute path, as step 4 does: the counters agree, the error counts are zero\n   and the work counts match, or stop and say which number differed. This is a stability check\n   only; its number is never compared at the end. Run the holdout nowhere else until the run ends.\n\n## One attempt\n\n1. **Read the log**: `python3 \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Fclimb\u002Fscripts\u002Fclimb_log.py\" show --log \u003Cslice>\u002Flog.tsv`.\n   Do not retry a mechanism already reverted unless the new attempt says what differs.\n2. **State one hypothesis that names a mechanism**: what the code does now, why that costs the\n   counter, and what the change removes. \"Templates are parsed once per page; parsing once per\n   build removes N-1 parses\" qualifies. \"The renderer does too much\" does not: it names no code\n   path and predicts no number.\n3. **Make one change** in `\u003Cclimb-path>` that tests that hypothesis and nothing else. Never edit\n   the harness files, the holdout harness included. Never read holdout output while forming a\n   hypothesis; holdout numbers never go into `log.tsv`, so `show` carries none.\n4. **Commit it** on `climb\u002F\u003Cgoal-slug>`, staging the changed files by name.\n5. **Run the project's tests** in `\u003Cclimb-path>`. Record `pass` or `fail`.\n6. **Measure both arms in one sitting.** Run the snapshot harness `ab.sh` with `--a` running the\n   harness in `\u003Cbaseline-path>` and `--b` running it in `\u003Cclimb-path>`, at the goal's iterations and\n   percentiles. `ab.sh` discards each arm's output, so also run the harness once in each worktree\n   and read its counter, error count and work count. Never compare against a number stored from an\n   earlier attempt; the baseline arm is re-measured every time. Never run the holdout harness here.\n7. **Decide**: `python3 \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Fclimb\u002Fscripts\u002Fclimb_log.py\" decide --rule \u003Cslice>\u002Fkeep-rule.json --before \u003Cbaseline counter> --after \u003Cattempt counter>`.\n   Settle the verdict in this order, stopping at the first line that applies:\n   1. Tests failed, the error count rose, or the work count differs from the baseline's:\n      `reverted`, whatever `decide` prints. Finishing fewer work units is not a speedup.\n   2. `decide` prints `kept`: `kept`.\n   3. `decide` prints `reverted`, the counter is no worse than the baseline's, and\n      `git diff --numstat \u003Clast kept sha> \u003Cattempt-sha>` shows lines removed and none added:\n      `kept`, noted `simplification`. Deleting code without losing ground on the counter is a\n      result in its own right.\n   4. Otherwise: `reverted`.\n8. **Keep or revert.**\n   - Kept: move the baseline arm to the attempt, `git -C \u003Cbaseline-path> checkout --detach \u003Cattempt-sha>`.\n   - Reverted: in `\u003Cclimb-path>`, `git reset --keep \u003Clast kept sha>`, which moves only climb's own\n     branch. If it refuses, stop and report the files; never use another reset form to force it.\n9. **Log the row**: `python3 \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Fclimb\u002Fscripts\u002Fclimb_log.py\" append --log \u003Cslice>\u002Flog.tsv`.\n   Every row passes all nine flags. The outcome goes in `--verdict` (`kept` or `reverted`) and\n   `--tests` (`pass` or `fail`); the measured numbers in `--before`, `--after` and `--delta`; the\n   text in `--id`, `--hypothesis`, `--change` and `--note`. When the verdict is `reverted`, put the\n   dropped SHA in `--note`; `git show` on that SHA still displays the attempt.\n\n- **Pointer**: for what `--keep` refuses and what it moves, see\n  [git-reset `--keep`](https:\u002F\u002Fgit-scm.com\u002Fdocs\u002Fgit-reset#Documentation\u002Fgit-reset.txt---keep).\n\n## When the run ends\n\n| Logged attempts | Last kept value | Budget | Untried mechanism | Action |\n|---|---|---|---|---|\n| any | any | spent | any | finish; report the target as not met when it is short |\n| at or above `min_attempts` | meets the Realistic target | left or not set | any | finish |\n| any | short of the target, or `min_attempts` not reached | left or not set | at least one | next attempt |\n| any | short of the target, or `min_attempts` not reached | left or not set | none | finish; report `Next idea: none left` and which condition was not reached |\n\nA budget is optional. Without one, the last row is what ends a run that never reaches its target.\n\n`min_attempts` is a floor on evidence: the attempts after the first kept one show whether the\ncounter has more room. The goal, the keep rule and the budget stay as written until the run ends.\n\n## Choosing the next hypothesis\n\n- **Smaller first.** Between two hypotheses expected to move the counter by about the same amount,\n  attempt the one with the smaller change first; the number never justifies extra code on its own.\n\n- **From the log.** Group the logged rows by the mechanism each one named. Prefer a mechanism no\n  row has tried yet over a variant of a reverted one.\n- **After three reverts in a row**, run `\u002Fperformance:target` against `\u003Cclimb-path>` when it is\n  among the available skills, and take the next hypothesis from its ranking; without it, profile\n  the harness run once and pick the largest cost no row has touched.\n- **Report what each gain cost.** Each kept line in the report carries its `git diff --stat`\n  totals, so the reviewer can weigh a small gain against a large change.\n\n## Unattended runs\n\nThe skill starts no `\u002Fgoal` and sets no loop. For a run nobody watches, draft its stop condition\nfirst with `\u002Fplanning:draft-goal-condition` when that skill is among the available skills,\nnaming the goal's target, `min_attempts` and the budget; without it, write the condition from\nthose same three items. The person starts the run with that condition.\n\n## Finish\n\n0. **Holdout, only when item 4 is given.** Both worktrees sit at the last kept sha. When it equals\n   the start sha, skip the measurement and report `Holdout:   not run (nothing kept)`. Otherwise\n   run `git -C \u003Cbaseline-path> checkout --detach \u003Cstart-sha>`, then run the holdout once in each\n   worktree, by absolute path, and read its counter, error count and work count; no `ab.sh` run\n   is needed. Call `decide --rule \u003Cslice>\u002Fkeep-rule.json --before \u003Choldout counter on the start sha>\n   --after \u003Choldout counter on the last kept sha>`: `kept` reports `held`, `reverted` reports\n   `did not hold`. As in step 7.1, a rose error count or a differing work count reports\n   `did not hold` whatever `decide` prints. A run whose kept attempts are all simplifications\n   (step 7.3) reports whatever `decide` prints; a `did not hold` there says only that the holdout\n   showed no gain. The holdout informs and never reverts: a gain that did not hold is still the\n   climb's result. Do this before step 1 removes the worktrees.\n1. Remove both worktrees with `git worktree remove \u003Cpath>`. The branch `climb\u002F\u003Cgoal-slug>` stays\n   and holds the kept commits in commit order. When `git worktree remove` refuses, leave that\n   worktree in place, name its path and the files `git status --porcelain` lists in the report, and\n   let the person clean it. Never pass `--force`.\n2. Report:\n\n```text\nBranch:    climb\u002F\u003Cgoal-slug>\nLog:       \u003Cabsolute path to log.tsv>\nCounter:   \u003Ccounter>, \u003Cdirection>; Realistic target \u003CRealistic target>\nMoved:     \u003Cstart value> -> \u003Clast kept value> (\u003Cpercent change>)\nEvidence:  min_attempts \u003Cn>; logged \u003Cn>: kept \u003Ck>, reverted \u003Cr>\nHoldout:   \u003Cheld|did not hold>; \u003Choldout counter on start sha> -> \u003Choldout counter on last kept sha>, or \"not run (nothing kept)\"   (only when a holdout was given; with errors or work counts that differ, name them)\nNext idea: \u003Cthe untried hypothesis most likely to move the counter, or \"none left\">\nChanges, oldest first, one line each:\n  \u003Csha> \u003Cchange> (\u003Cfiles> files, +\u003Cadded> -\u003Cremoved>)\n```\n\n- **Pointer**: for which worktrees `remove` accepts, see\n  [git-worktree `remove`](https:\u002F\u002Fgit-scm.com\u002Fdocs\u002Fgit-worktree#Documentation\u002Fgit-worktree.txt-remove).\n  Ignored-only output was also checked by hand: removal succeeds.\n\n## Boundary\n\n- **Does not set the goal or the rule's counter.** `\u002Fperformance:goal` does, with a person.\n- **Does not freeze the harness.** `\u002Fperformance:snapshot` does.\n- **Does not decide the result is real.** `\u002Fperformance:verify` re-derives it in a fresh context.\n- **Does not climb an eval score.** For an LLM app scored against an eval set, the person types\n  `\u002Fclaude-api hillclimb`; this skill never starts it.\n\n- **Pointer**: for the `\u002Fclaude-api` subcommands, see\n  \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fskills#work-on-claude-api-projects>. **As of**: 2026-10-04.\n  **Recheck trigger**: that section drops or renames `hillclimb`.\n\n## Next\n\n`\u002Fperformance:verify` on the branch `climb\u002F\u003Cgoal-slug>`, with the log path.\n\n## Gotchas\n\n- **The rule file judges every gain.** If an attempt looks good and `decide` says `reverted`, the\n  attempt is reverted unless it is a pure deletion (step 7, line 3). Change the goal through\n  `\u002Fperformance:goal` before the next run, never during this one.\n- **A counter that stops moving can mean the harness stopped running the code.** Check the work\n  count before concluding a mechanism had no effect.\n- **The baseline worktree is detached on purpose.** It follows kept commits only; a branch there\n  would collide with `climb\u002F\u003Cgoal-slug>`.\n- **Run the harness, and the holdout harness, by absolute path in both worktrees.** A relative\n  path run from the wrong directory measures the same tree twice.\n","**Arguments.** `[\u003Cgoal-slug>]`. e.g. \u002Fperformance:climb site-build-template-parses\n\n## Purpose\n\nAnswers **\"which of these changes actually moved the number, and how far can it go?\"**\n\nMeasure one change at a time: hypothesis, change, commit, measure both arms, apply the rule, keep\nor revert, log, until the goal is met. Keep every judgment out of reach of the attempt being\njudged: the rule is written first, the harness is frozen, and the baseline is re-measured beside\neach attempt.\n\nRead [`${CLAUDE_PLUGIN_ROOT}\u002Freference\u002Fharness-integrity.md`](${CLAUDE_PLUGIN_ROOT}\u002Freference\u002Fharness-integrity.md)\nbefore the first attempt.\n\n## Inputs, or stop\n\n1. **A goal from `\u002Fperformance:goal`** with its floor, its Realistic target and `min_attempts`. A\n   goal without `min_attempts` goes back to `\u002Fperformance:goal`; never pick one here, since goal is\n   human-gated.\n2. **A harness `\u002Fperformance:snapshot` froze**: the command, the commit it was frozen at, and its\n   discrimination result. Send the run back to `\u002Fperformance:snapshot baseline` when no harness is\n   frozen, or when its output carries neither tally. Each harness run must print three numbers: the\n   goal's counter, a tally of failures and a tally of finished work units.\n3. **A clean checkout.** `git status --porcelain` in your checkout prints nothing. Refuse\n   otherwise and name the files: the loop starts from a commit, and uncommitted work is not in it.\n4. **Optional: a holdout harness `\u002Fperformance:snapshot` froze**, a second command that prints the\n   same three numbers on inputs the main harness does not use. It needs its own `Frozen:` line in\n   the baseline report with `discriminate: exit 0`, worded like item 2; send the run back to\n   `\u002Fperformance:snapshot baseline` when the line is missing. **Absent means off**: the run behaves\n   exactly as it does without this item, and nothing below that mentions the holdout applies.\n\n## Set up, before the first attempt\n\n1. **Memory slice.** Use `\u003Cmemory_dir>\u002F\u003Ctopic-slug>\u002Fclimb\u002F\u003Cgoal-slug>\u002F` in your checkout (default\n   `\u003Cmemory_dir>` is `.work\u002F`), by absolute path: the loop runs in another worktree, and the log\n   must outlive it. Done when the directory exists.\n2. **Keep rule.** Write `keep-rule.json` there with exactly three keys: `counter` (the goal's\n   counter, by name), `direction` (`lower` or `higher`) and `gain_floor` (the change in the counter\n   that counts as a gain; the host's measured noise for a duration, `0` for a deterministic count).\n   Show it to the person when one is present. **Never edit it during the run**: once an attempt\n   has a number, any change to the file would tailor the rule to that number.\n3. **Two worktrees, outside your checkout or in a directory its `.gitignore` ignores.** Start both\n   at your checkout's `HEAD`:\n   - climb's own: `git worktree add -b climb\u002F\u003Cgoal-slug> \u003Cclimb-path> \u003Cstart-sha>`. A branch with\n     that name already exists: stop and ask, since it may hold an earlier run.\n   - the baseline arm: `git worktree add --detach \u003Cbaseline-path> \u003Cstart-sha>`. It always sits at\n     the last kept commit.\n\n   Every attempt happens in `\u003Cclimb-path>`. Nothing in this skill edits, commits in or resets your\n   checkout.\n4. **Starting measurement.** Run the harness once in each worktree. The counters agree, the error\n   counts are zero and the work counts match, or the harness is not stable enough to climb on:\n   stop and say which number differed.\n5. **Holdout, only when item 4 is given.** Write `holdout.json` in the slice, holding the\n   holdout's `Frozen:` command and sha; like `keep-rule.json` it is never edited during the run.\n   It is not a key of `keep-rule.json`, which `decide` accepts only with its three keys. Then run\n   the holdout once in each worktree, by absolute path, as step 4 does: the counters agree, the error counts are zero\n   and the work counts match, or stop and say which number differed. This is a stability check\n   only; its number is never compared at the end. Run the holdout nowhere else until the run ends.\n\n## One attempt\n\n1. **Read the log**: `python3 \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Fclimb\u002Fscripts\u002Fclimb_log.py\" show --log \u003Cslice>\u002Flog.tsv`.\n   Do not retry a mechanism already reverted unless the new attempt says what differs.\n2. **State one hypothesis that names a mechanism**: what the code does now, why that costs the\n   counter, and what the change removes. \"Templates are parsed once per page; parsing once per\n   build removes N-1 parses\" qualifies. \"The renderer does too much\" does not: it names no code\n   path and predicts no number.\n3. **Make one change** in `\u003Cclimb-path>` that tests that hypothesis and nothing else. Never edit\n   the harness files, the holdout harness included. Never read holdout output while forming a\n   hypothesis; holdout numbers never go into `log.tsv`, so `show` carries none.\n4. **Commit it** on `climb\u002F\u003Cgoal-slug>`, staging the changed files by name.\n5. **Run the project's tests** in `\u003Cclimb-path>`. Record `pass` or `fail`.\n6. **Measure both arms in one sitting.** Run the snapshot harness `ab.sh` with `--a` running the\n   harness in `\u003Cbaseline-path>` and `--b` running it in `\u003Cclimb-path>`, at the goal's iterations and\n   percentiles. `ab.sh` discards each arm's output, so also run the harness once in each worktree\n   and read its counter, error count and work count. Never compare against a number stored from an\n   earlier attempt; the baseline arm is re-measured every time. Never run the holdout harness here.\n7. **Decide**: `python3 \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Fclimb\u002Fscripts\u002Fclimb_log.py\" decide --rule \u003Cslice>\u002Fkeep-rule.json --before \u003Cbaseline counter> --after \u003Cattempt counter>`.\n   Settle the verdict in this order, stopping at the first line that applies:\n   1. Tests failed, the error count rose, or the work count differs from the baseline's:\n      `reverted`, whatever `decide` prints. Finishing fewer work units is not a speedup.\n   2. `decide` prints `kept`: `kept`.\n   3. `decide` prints `reverted`, the counter is no worse than the baseline's, and\n      `git diff --numstat \u003Clast kept sha> \u003Cattempt-sha>` shows lines removed and none added:\n      `kept`, noted `simplification`. Deleting code without losing ground on the counter is a\n      result in its own right.\n   4. Otherwise: `reverted`.\n8. **Keep or revert.**\n   - Kept: move the baseline arm to the attempt, `git -C \u003Cbaseline-path> checkout --detach \u003Cattempt-sha>`.\n   - Reverted: in `\u003Cclimb-path>`, `git reset --keep \u003Clast kept sha>`, which moves only climb's own\n     branch. If it refuses, stop and report the files; never use another reset form to force it.\n9. **Log the row**: `python3 \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Fclimb\u002Fscripts\u002Fclimb_log.py\" append --log \u003Cslice>\u002Flog.tsv`.\n   Every row passes all nine flags. The outcome goes in `--verdict` (`kept` or `reverted`) and\n   `--tests` (`pass` or `fail`); the measured numbers in `--before`, `--after` and `--delta`; the\n   text in `--id`, `--hypothesis`, `--change` and `--note`. When the verdict is `reverted`, put the\n   dropped SHA in `--note`; `git show` on that SHA still displays the attempt.\n\n- **Pointer**: for what `--keep` refuses and what it moves, see\n  [git-reset `--keep`](https:\u002F\u002Fgit-scm.com\u002Fdocs\u002Fgit-reset#Documentation\u002Fgit-reset.txt---keep).\n\n## When the run ends\n\n| Logged attempts | Last kept value | Budget | Untried mechanism | Action |\n|---|---|---|---|---|\n| any | any | spent | any | finish; report the target as not met when it is short |\n| at or above `min_attempts` | meets the Realistic target | left or not set | any | finish |\n| any | short of the target, or `min_attempts` not reached | left or not set | at least one | next attempt |\n| any | short of the target, or `min_attempts` not reached | left or not set | none | finish; report `Next idea: none left` and which condition was not reached |\n\nA budget is optional. Without one, the last row is what ends a run that never reaches its target.\n\n`min_attempts` is a floor on evidence: the attempts after the first kept one show whether the\ncounter has more room. The goal, the keep rule and the budget stay as written until the run ends.\n\n## Choosing the next hypothesis\n\n- **Smaller first.** Between two hypotheses expected to move the counter by about the same amount,\n  attempt the one with the smaller change first; the number never justifies extra code on its own.\n\n- **From the log.** Group the logged rows by the mechanism each one named. Prefer a mechanism no\n  row has tried yet over a variant of a reverted one.\n- **After three reverts in a row**, run `\u002Fperformance:target` against `\u003Cclimb-path>` when it is\n  among the available skills, and take the next hypothesis from its ranking; without it, profile\n  the harness run once and pick the largest cost no row has touched.\n- **Report what each gain cost.** Each kept line in the report carries its `git diff --stat`\n  totals, so the reviewer can weigh a small gain against a large change.\n\n## Unattended runs\n\nThe skill starts no `\u002Fgoal` and sets no loop. For a run nobody watches, draft its stop condition\nfirst with `\u002Fplanning:draft-goal-condition` when that skill is among the available skills,\nnaming the goal's target, `min_attempts` and the budget; without it, write the condition from\nthose same three items. The person starts the run with that condition.\n\n## Finish\n\n0. **Holdout, only when item 4 is given.** Both worktrees sit at the last kept sha. When it equals\n   the start sha, skip the measurement and report `Holdout:   not run (nothing kept)`. Otherwise\n   run `git -C \u003Cbaseline-path> checkout --detach \u003Cstart-sha>`, then run the holdout once in each\n   worktree, by absolute path, and read its counter, error count and work count; no `ab.sh` run\n   is needed. Call `decide --rule \u003Cslice>\u002Fkeep-rule.json --before \u003Choldout counter on the start sha>\n   --after \u003Choldout counter on the last kept sha>`: `kept` reports `held`, `reverted` reports\n   `did not hold`. As in step 7.1, a rose error count or a differing work count reports\n   `did not hold` whatever `decide` prints. A run whose kept attempts are all simplifications\n   (step 7.3) reports whatever `decide` prints; a `did not hold` there says only that the holdout\n   showed no gain. The holdout informs and never reverts: a gain that did not hold is still the\n   climb's result. Do this before step 1 removes the worktrees.\n1. Remove both worktrees with `git worktree remove \u003Cpath>`. The branch `climb\u002F\u003Cgoal-slug>` stays\n   and holds the kept commits in commit order. When `git worktree remove` refuses, leave that\n   worktree in place, name its path and the files `git status --porcelain` lists in the report, and\n   let the person clean it. Never pass `--force`.\n2. Report:\n\n```text\nBranch:    climb\u002F\u003Cgoal-slug>\nLog:       \u003Cabsolute path to log.tsv>\nCounter:   \u003Ccounter>, \u003Cdirection>; Realistic target \u003CRealistic target>\nMoved:     \u003Cstart value> -> \u003Clast kept value> (\u003Cpercent change>)\nEvidence:  min_attempts \u003Cn>; logged \u003Cn>: kept \u003Ck>, reverted \u003Cr>\nHoldout:   \u003Cheld|did not hold>; \u003Choldout counter on start sha> -> \u003Choldout counter on last kept sha>, or \"not run (nothing kept)\"   (only when a holdout was given; with errors or work counts that differ, name them)\nNext idea: \u003Cthe untried hypothesis most likely to move the counter, or \"none left\">\nChanges, oldest first, one line each:\n  \u003Csha> \u003Cchange> (\u003Cfiles> files, +\u003Cadded> -\u003Cremoved>)\n```\n\n- **Pointer**: for which worktrees `remove` accepts, see\n  [git-worktree `remove`](https:\u002F\u002Fgit-scm.com\u002Fdocs\u002Fgit-worktree#Documentation\u002Fgit-worktree.txt-remove).\n  Ignored-only output was also checked by hand: removal succeeds.\n\n## Boundary\n\n- **Does not set the goal or the rule's counter.** `\u002Fperformance:goal` does, with a person.\n- **Does not freeze the harness.** `\u002Fperformance:snapshot` does.\n- **Does not decide the result is real.** `\u002Fperformance:verify` re-derives it in a fresh context.\n- **Does not climb an eval score.** For an LLM app scored against an eval set, the person types\n  `\u002Fclaude-api hillclimb`; this skill never starts it.\n\n- **Pointer**: for the `\u002Fclaude-api` subcommands, see\n  \u003Chttps:\u002F\u002Fcode.claude.com\u002Fdocs\u002Fen\u002Fskills#work-on-claude-api-projects>. **As of**: 2026-10-04.\n  **Recheck trigger**: that section drops or renames `hillclimb`.\n\n## Next\n\n`\u002Fperformance:verify` on the branch `climb\u002F\u003Cgoal-slug>`, with the log path.\n\n## Gotchas\n\n- **The rule file judges every gain.** If an attempt looks good and `decide` says `reverted`, the\n  attempt is reverted unless it is a pure deletion (step 7, line 3). Change the goal through\n  `\u002Fperformance:goal` before the next run, never during this one.\n- **A counter that stops moving can mean the harness stopped running the code.** Check the work\n  count before concluding a mechanism had no effect.\n- **The baseline worktree is detached on purpose.** It follows kept commits only; a branch there\n  would collide with `climb\u002F\u003Cgoal-slug>`.\n- **Run the harness, and the holdout harness, by absolute path in both worktrees.** A relative\n  path run from the wrong directory measures the same tree twice.\n",{"metadata":24,"description":7,"argument-hint":27,"user-invocable":28,"disable-model-invocation":20},{"summary":25,"workflow-stage":26},"Keep or revert one measured change at a time until the goal is met","implement","[\u003Cgoal-slug>]",true,"main","d0e68c3f36b92a1d6381f0ba0060d7a0b8b886c5","plugins\u002Fperformance\u002Fskills\u002Fclimb",7,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins","https:\u002F\u002Fgithub.com\u002Fmelodic-software","https:\u002F\u002Fmelodicsoftware.com","Independent software studio building developer tools, automation, and application services, with roots in music.","2026-11-15 05:15:06",{"source":8,"skillName":5,"commands":39,"prompt":49,"zipAvailable":28},[40,43,46],{"tool":41,"command":42},"npx","npx skills add melodic-software\u002Fclaude-code-plugins --skill climb",{"tool":44,"command":45},"bunx","bunx skills add melodic-software\u002Fclaude-code-plugins --skill climb",{"tool":47,"command":48},"pnpm","pnpm dlx skills add melodic-software\u002Fclaude-code-plugins --skill climb","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fperformance\u002Fskills\u002Fclimb\u002FSKILL.md",[51,57,63],{"path":52,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":55,"sourceHtmlUrl":56},"evals\u002Fevals.json","REFERENCE","参考文档",8739,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fperformance\u002Fskills\u002Fclimb\u002Fevals\u002Fevals.json",{"path":58,"resourceType":59,"resourceTypeDesc":60,"sizeBytes":61,"sourceHtmlUrl":62},"scripts\u002Fclimb_log.py","SCRIPT","脚本",5542,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fperformance\u002Fskills\u002Fclimb\u002Fscripts\u002Fclimb_log.py",{"path":64,"resourceType":59,"resourceTypeDesc":60,"sizeBytes":65,"sourceHtmlUrl":66},"scripts\u002Ftest_climb_log.py",14023,"https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fperformance\u002Fskills\u002Fclimb\u002Fscripts\u002Ftest_climb_log.py",[68,75,82,89,94,101],{"skillNo":69,"name":70,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":72,"categories":73,"keywords":74,"publishedAt":19},"01M7DYYDG31B9YX7N9MB9QW2GW","track","Backlog reads through the bound tracker: search open and closed items, view one item with its comments and state history. Use when another skill searches or reads tracker items.","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fsession-flow\u002Fevals\u002Frecall-still-reports-reopened-bug\u002Fwork-items-stub\u002Fskills\u002Ftrack\u002FSKILL.md",[],[],{"skillNo":76,"name":77,"nameZh":6,"description":78,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":79,"categories":80,"keywords":81,"publishedAt":19},"01M7DYXVBAVG3NV30PA7GKW9Q6","reviewer-media","Reviewer media for one pull request: decides whether it needs a narrated walkthrough, builds the storyboard from \u002Freview:explain-change's digest, has \u002Fexplainer-video:produce render it and hands the manifest to \u002Fsource-control:pull-request media. Use when asked for 'reviewer media', 'a walkthrough video for this PR', 'a produced video of this change', or when pull-request ready or monitor hands a PR over.","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Freview\u002Fskills\u002Freviewer-media\u002FSKILL.md",[],[],{"skillNo":83,"name":84,"nameZh":6,"description":85,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":86,"categories":87,"keywords":88,"publishedAt":19},"01M7DYRZ5YD44AMYH26YR25GZ8","route","Model and effort per agent role (orchestrator, worker, verifier, retrieval) for a multi-agent run, resolved through the bundled defaults and the user, team and personal override config layers into fenced JSON with single-agent and fan-out variants and the layer behind each value. Use when asked 'which model for the subagents', 'route the roles' or 'model routing', or before a skill passes a role's model on a spawn.","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fdiscovery\u002Fevals\u002Fex-model-pin-without-override\u002Fmulti-agent-stub\u002Fskills\u002Froute\u002FSKILL.md",[],[],{"skillNo":90,"name":84,"nameZh":6,"description":85,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":91,"categories":92,"keywords":93,"publishedAt":19},"01M7DYRTVT4FWTGDJY0HCCFJMZ","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Fdiscovery\u002Fevals\u002Fex-model-override-from-route\u002Fmulti-agent-stub\u002Fskills\u002Froute\u002FSKILL.md",[],[],{"skillNo":95,"name":96,"nameZh":6,"description":97,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":98,"categories":99,"keywords":100,"publishedAt":19},"01M7DYNZC8PNW3T27V2ADKECKJ","principles","Module-design vocabulary for architecture findings, and the checks built on it. Use when: 'what is a deep module', 'what is a seam', 'deep vs shallow module', 'what is the deletion test', 'leverage and locality', 'module design vocabulary', 'which word do we use for X in architecture findings'. Scanning a codebase for friction is \u002Farchitecture:improve.","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Farchitecture\u002Fskills\u002Fprinciples\u002FSKILL.md",[],[],{"skillNo":102,"name":103,"nameZh":6,"description":104,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":105,"categories":106,"keywords":107,"publishedAt":19},"01M7DYNMCZNEWBFHV9KTSS6H60","write","Structured five-field bug report (title, steps to reproduce, expected vs actual, justified severity, suggested fix location) from an informal description, read-only. Use when the user names a defect they observed, or asks for one written up ('report a bug', 'file a bug', 'write this up as a bug').","https:\u002F\u002Fgithub.com\u002Fmelodic-software\u002Fclaude-code-plugins\u002Fblob\u002Fmain\u002Fplugins\u002Farchitecture\u002Fevals\u002Fdeepening-defect-goes-to-bugs-write\u002Fbugs-stub\u002Fskills\u002Fwrite\u002FSKILL.md",[],[]]