Hill Climb
Turn a fuzzy goal into a score, then climb the score. One small change at a
time, keep it only if the score goes up, revert if it goes down. Stop at the
threshold, at a plateau, or at the round cap — whichever comes first.
Origin: Theo Browne (t3.gg), "Agents Sucked At UI (Until Today)", 2026-11-03,
and the council skill in t3dotgg/tstack. Theo's version scores UI with
three models from three providers. This skill generalises it: any goal that
can be scored, with one implementer provider, one model-judge provider,
and one deterministic gate.
When To Use
- The goal has no single right answer, but quality can be scored 0-10:
code quality, a research idea, a UI, docs, a prompt, a design.
- The user says "hill climb", "keep improving until it scores X", or
"iterate until the judges are happy".
- A deterministic check exists (tests, build, lint, a checklist, counts)
that can act as a hard gate alongside model judges.
When NOT To Use
- One-pass edits, routine fixes, or a task with a binary pass/fail and no
quality gradient — use
bounded-agent-loop instead.
- No score can be defined even after the walkthrough below. If you cannot
say what "better" means and how to measure it, stop and ask. No score,
no hill. Do not guess an eval silently.
- Unattended scheduling. A hill climb is a supervised, bounded run, not a
cron job. Do not merge, deploy, or enable a schedule from it.
Judges
Only two model providers are assumed available. Do not invent a third.
| Seat |
Provider |
Role |
| Implementer, first |
agy (Gemini) |
Makes the change |
| Model judge / critic |
Codex (OpenAI) |
Scores it, lists fixes, never implements its own review |
| 3rd judge — deterministic |
Tests / build / lint / checklist / counts |
Hard gate. Must pass before any model score counts. No model, no cost, no bias. |
Full Mode has ONE model judge (Codex), not two. The deterministic gate
is the 3rd judge, pass/fail, never scored. There is no second model
judge in Full Mode — the implementer never scores its own work.
Do not assume any other provider is available. Check before using one;
if it is unavailable, report it instead of substituting. A third model
may be used only as a final tie-breaker if the user explicitly asks,
and say so in the report.
Swap implementer and critic on the next run if the user asks, or if one
model stalls. Never let the same model review its own work in the same
round: same family is the same person at a different time of day.
Exception: the explicitly labelled degraded 1-model and no-delegate
modes below, where the same model reviews in a fresh context and the
score is labelled single-judge / self-review, low confidence, never
consensus. That exception does not apply in Full Mode.
The orchestrator is not a judge in this loop. It drives the agent pane,
reads the result, and decides keep / revert / next step. The user
is the final judge on taste calls.
If the Harness Cannot Call the Other Models (Fallback)
Check availability BEFORE starting the loop, not mid-climb: agy auth,
Codex auth / quota, agent pane reachable, deterministic gate command
exists and runs. Report exactly what is missing. Never invent a
provider, never fake a score, never silently substitute a same-family
model and call it a different judge.
| Mode |
When |
What changes |
| Full |
agy + Codex + deterministic gate all reachable |
As in The Loop below. Threshold = model judge >= 9 AND gate passed. |
| Degraded, 1 model |
Only one of agy / Codex reachable |
Implementer = available model. Critic = same model in a fresh context, different role prompt, read-only. Deterministic gate still hard. Human is final judge. Label every score "single-judge, low confidence". A 9/10 here is NOT consensus — if it hits 9, pause and ask the user whether to accept it as final. Cap at 4 rounds unless the user says more. |
| Degraded, 0 models |
No other model reachable, only the orchestrator |
No model scores at all. Deterministic gate only. Do not invent scores. Run as bounded-agent-loop instead, or stop and ask the user to score it / supply a judge. |
| No delegate tool |
Harness cannot spawn subagents / agent pane unavailable |
Run sequentially in one pane, one round at a time, with the review in a FRESH context (new session / fresh turn, never the same turn as the implementation). This is single-agent self-review, NOT a second judge: label every score "self-review, low confidence", human is final judge, cap at 4 rounds. If it hits 9, pause and ask the user whether to accept it as final. Threshold here does NOT count as consensus. See requesting-code-review fallback. |
Rules for any degraded mode:
- Tell the user the mode, which seats are missing, and what that costs
in confidence — then get their "go" before continuing, unless they
pre-approved degraded mode for this run.
- Never run at API prices without warning the user first.
- If a provider dies mid-climb: checkpoint the best version, stop the
round, report what was lost, and ask whether to continue degraded,
swap implementer / critic, or stop. Do not carry on as if nothing
happened.
If the Goal / Eval Is Unspecified (Walkthrough)
Do NOT start the loop with a blank goal or eval. Walk the user through
it first. Ask at most 5 questions, one at a time, in plain language.
Suggest a default with every question, so the user can just say "use
that", edit it, or say "I don't know".
- What are we improving, in one sentence?
- What does good look like — what would a 9/10 mean here?
- How do we check it — tests, a checklist, screenshots, sources?
- What must NOT change?
- When do we stop — threshold, rounds, or time?
If the user is unsure, do the suggesting for them. Concrete defaults
per question — offer these verbatim, do not invent different ones:
| # |
Question |
Suggested default |
| 1 |
What are we improving? |
"Improve <the artefact, in the user's words> until the judges score it 9/10" |
| 2 |
What does 9/10 mean? |
The Model score column from the Define the Eval table for their task type (coding / research / UI / docs) |
| 3 |
How do we check it? |
The Deterministic gate column from the same table (tests + lint + build; sources checked; screenshots; checklist) |
| 4 |
What must NOT change? |
Tests, the eval itself, and every file in Immutable — by default, nothing outside Editable files |
| 5 |
When do we stop? |
Threshold 9/10 from the model judge, gate passed, max 8 rounds, plateau 2 rounds with no gain |
- Propose a draft goal and eval from the Define the Eval table for
their task type (coding / research / UI / docs), with defaults:
threshold 9/10, max 8 rounds, plateau 2.
- If they say "I don't know" to a question, offer 2-3 options,
recommend one, and say why in one line. Let them pick.
- Fill in the authorization form below with your draft, show it back,
and get an explicit "go" before starting. The user never has to fill
a blank form.
- Score the baseline first and show it to them. Seeing a 5/10 and what
the judges would fix often clarifies the goal faster than more
questions.
If after the walkthrough there is still no scoreable eval, stop. Do
not guess one silently. Say what is missing, and offer bounded-agent-loop
or a one-pass draft instead.
Authorizing a Run
Needs explicit authorization and hard limits before it starts. Fill every
field — an unfilled budget is not an unlimited budget, it is a reason to
stop and ask.
Run a hill climb.
Goal: <one sentence, what "top of the hill" looks like>
Eval: <how to score 0-10, who scores, what counts>
Deterministic gate: <exact command / checklist that must pass first>
Threshold: <default 9/10 from the model judge, gate passed>
Max rounds: <default 8>
Plateau: <default 2 rounds with no gain = stop>
Baseline: <score the starting version — must be scored before the loop starts>
Editable files: <explicit paths the worker may change>
Immutable: <tests, eval, and files it must not touch>
Worktree/branch: <disposable branch or worktree to use>
Cost cap: <limit, or "none enforceable">
Usage pacing: <budget per hour as a share of the ACTIVE provider's
limit window, e.g. ~20% per hour; never 100% — a run
that burns the window stalls. The window belongs to
whichever agent is running (Codex, agy, OpenCode,
Muse) — agy's is 5h; use the active provider's own
window. Check remaining budget at every round
boundary; slow down or pause before the window is
exhausted>
Define the Eval
| Task |
Deterministic gate |
Model score 0-10 |
| Coding |
tests + lint + build pass, no regressions |
Correctness, clarity, scope discipline, no over-engineering |
| Research idea |
Sources checked, claims traceable, gaps listed |
Accuracy, evidence, coverage, novelty, falsifiability |
| UI |
Screenshots light + dark, desktop + phone, empty + full |
Layout, hierarchy, consistency, distinctiveness |
| Docs / writing |
Checklist: facts verified, no invented claims |
Clarity, completeness, tone, structure |
Score the baseline first, before any change — in this order: run the
deterministic gate on the baseline, then score it. If the gate fails
on the baseline, the baseline score is reference-only and the loop
does NOT start until the gate passes; fix the baseline first. A
gate-failed baseline score never counts as a kept score.
Every 3rd round, re-score the baseline: if the re-score differs by
more than 1 from the original baseline, report the drift and re-anchor
like this: set baseline = new re-score, set best kept score = new
re-score, and mark all earlier scores as old-scale in the report —
pre-drift and post-drift scores are not comparable. Calibration: models
rate themselves too high. When the user scores the artefact, log the gap
between the model judge's score and the user's score as the calibration
offset, and subtract it from every later model score. Recompute the
offset each time the user scores again; the latest offset replaces the
old one. If a
judge scores something that is not in the screenshots / code / sources
in front of it, discard that finding and say so — check every finding
against the artefact before acting on it.
The Loop
Score rule — read this before step 5. There is ONE model judge and
one deterministic gate, and "the score went up" means all of this:
- Score = the model judge's 0-10 score for that round. The
deterministic gate is pass/fail, it is never scored.
- Keep = judge score rose vs the best kept score so far, AND the gate
passed. Compare only against the last kept round or the baseline —
never against a reverted round.
- Threshold = model judge scores >= 9 AND the gate passes in the same
round.
- Plateau = judge score did not rise above the best kept score for 2
consecutive rounds. Flat is not a gain.
Checkpoint / revert mechanics. The loop runs ONLY on a disposable
branch / worktree. Before starting, verify the worktree is clean
(git status --porcelain empty, except files the loop is about to
edit). Stage ONLY the editable files from the authorization form —
never the whole tree: git add -- <editable files>. Checkpoint the
baseline first: commit those files with git commit -m "hill-climb baseline". Record that commit as the best checkpoint. On every kept
round, commit the editable files again:
git commit -m "hill-climb round N: <score>/10", and update the best
checkpoint. On every revert, go back to the best checkpoint with
git restore -- <editable files> plus
git clean -fd -- <editable files> (or git reset --hard <best-checkpoint> as a last resort, disposable branch only). These
commands destroy uncommitted state — never run them on a shared
branch, and never run git add -A or a bare git clean -fd here.
Never let dirty edits from a reverted round contaminate the next
attempt.
- Record goal, eval, threshold, max rounds, plateau, baseline score,
editable files, and immutable eval. Checkpoint the baseline
(commit on the disposable branch, per above).
- Implementer (agy first) makes ONE small change toward the judges'
last notes. Small, or the score change tells you nothing.
- Run the deterministic gate. Fail = revert (per above), record the
failure, next attempt. Never weaken the gate, the tests, or the eval
to get a pass.
- The model judge scores the new version 0-10, read-only, with
specific fixes listed. The judge sees only the artefact and the eval
rubric — no implementation history, no earlier rounds, no notes on
what changed. A judge that knows what was attempted scores the
attempt, not the artefact. A fresh review after fixes is a new round.
Every 3rd round, re-score the baseline (see Define the Eval); if
drift is more than 1, report it and re-anchor.
- Keep only if the judge score went up vs the best kept score (see
Score rule above) AND the gate passed. On keep, commit and update
the best checkpoint. On revert, reset to the best checkpoint.
Revert if it went down or flat with no new information. Keep the
best version seen so far, always.
- Stop on: threshold reached by the model judge with the gate passed,
plateau (default 2 rounds with no gain, per Score rule), max rounds,
budget out, scope change, or missing tools / credentials. Plateau is
a decision point, not a failure — report the best score and ask the
user what the top is.
Theo's line: "once you're climbing the hill, you decide what the top is."
- Report rounds in a table, best score, what changed, what stalled,
and the branch / checkpoint. Do not merge or deploy. That is a
separate authorization.
Escaping a Local Peak
Hill climbing gets stuck. When the score is flat for 2 rounds, the
loop STOPS (plateau). Do not autonomously branch, restructure, or
keep iterating. Present these options to the user and wait for their
pick — only proceed on their explicit pick:
- Change the approach, not the colour: restructure, split, remove, or
restart from the best checkpoint with a different strategy.
- Or branch 2-3 different approaches in parallel worktrees, score each
once, keep the best branch.
- Or stop. A flat score is usually waiting on a human decision, not
more loops.
Scope
Fight scope creep. The judges will find more things every round; fix
only what moves the eval score for this goal. Park the rest as
follow-ups. A climb that keeps widening never reaches the top.
Agent Delegation
- Work through a visible agent pane, one step at a time, pasted so it
is visible. Read the result, evaluate it, then paste the next step
or a correction without asking. Report only when DONE, BLOCKED, or
needing a decision.
- Never use headless exec (
codex exec, opencode run with a piped
prompt) for fleet-machine work — use a visible pane. On a standalone
machine with no agent pane, one headless run is allowed for a
review / audit step only: prompt from a file, log to the agent's log
directory, no edits beyond the report. Never hijack another live agent.
- The orchestrator drives; the agent in the pane does the work.
Debian / Linux (Standalone) — Addendum
The shared skill-authoring / skill-create texts are Windows /
OneDrive-first. Linux / Debian is first-class here too.
- Resolve the library clone from
~/.config/theprawnskills/library.json
(library key), never guess the path.
- Installed skills:
~/.agents/skills (Linux, symlinks). Never edit
an installed copy; edit the clone.
- Installer (Python 3.11+): run the repository installer from the library
root — see
platforms/linux/README.md for the exact
--dry-run / --apply / --check commands. Auto mode is symlinks on
Linux, copies on Windows.
- Shell: bash with
python3 on Linux, PowerShell on Windows.
Do not copy Windows paths, OneDrive paths, or dotagents sync
into a Linux run. Do not use the Windows machine registry
for standalone propagation.
- This skill is ON-DEMAND, not in
default-profile.toml, unless
daily exposure is requested. Keep default-profile.toml as the
shared daily selection; check all supported OSes before changing
profile membership.
Related Skills
bounded-agent-loop — the closest existing skill; bounded repair
loop with an acceptance command, max 3 attempts. Use it when the
eval is pass/fail, not a score.
skill-judge — multi-dimensional scoring; the scoring half of a
council, for skills.
requesting-code-review — review, fix, re-review loop.
skill-create, skill-authoring — how this skill was made.
Standalone variants live under platforms/linux/skills/.
skill-creator — generic upstream skill-creation loop
(draft, test, eval, rewrite, repeat). Method only; its Claude-only
eval tooling and paths do not apply here.
skill-add — import an existing skill from GitHub / a link / pasted
text. Not used here: this skill is new, not imported, and Theo's
council was adapted, not copied verbatim.
Provenance
Adapted from Theo Browne's council skill (t3dotgg/tstack, MIT) and
the video "Agents Sucked At UI (Until Today)", 2026-11-03. Generalised
from UI-only to any scoreable goal, and from three providers to one
implementer provider, one model-judge provider, and one deterministic
gate.
1---2name: hill-climb3description: >-4 Hill climb any scoreable goal — code, research, or UI — by looping implement, judge, fix until a threshold or plateau. Use when the user says hill climb, hill climbing, climb the hill, iterate until it scores, or keep improving until it hits 9, or when they ask the agent to walk them through defining the goal and eval first. Generic, not UI-only.5license: MIT6metadata:7 source: https://skillmd.com/skills/hongyime/hill-climb8 author: Local setup9 version: "1.5.0"10 platform: "Codex, Antigravity (agy), OpenCode, Muse on Windows, macOS, Linux (Debian)"11---1213# Hill Climb1415Turn a fuzzy goal into a score, then climb the score. One small change at a16time, keep it only if the score goes up, revert if it goes down. Stop at the17threshold, at a plateau, or at the round cap — whichever comes first.1819Origin: Theo Browne (t3.gg), "Agents Sucked At UI (Until Today)", 2026-11-03,20and the `council` skill in t3dotgg/tstack. Theo's version scores UI with21three models from three providers. This skill generalises it: any goal that22can be scored, with one implementer provider, one model-judge provider,23and one deterministic gate.2425## When To Use2627- The goal has no single right answer, but quality can be scored 0-10:28 code quality, a research idea, a UI, docs, a prompt, a design.29- The user says "hill climb", "keep improving until it scores X", or30 "iterate until the judges are happy".31- A deterministic check exists (tests, build, lint, a checklist, counts)32 that can act as a hard gate alongside model judges.3334## When NOT To Use3536- One-pass edits, routine fixes, or a task with a binary pass/fail and no37 quality gradient — use `bounded-agent-loop` instead.38- No score can be defined even after the walkthrough below. If you cannot39 say what "better" means and how to measure it, stop and ask. No score,40 no hill. Do not guess an eval silently.41- Unattended scheduling. A hill climb is a supervised, bounded run, not a42 cron job. Do not merge, deploy, or enable a schedule from it.4344## Judges4546Only two model providers are assumed available. Do not invent a third.4748| Seat | Provider | Role |49|---|---|---|50| Implementer, first | agy (Gemini) | Makes the change |51| Model judge / critic | Codex (OpenAI) | Scores it, lists fixes, never implements its own review |52| 3rd judge — deterministic | Tests / build / lint / checklist / counts | Hard gate. Must pass before any model score counts. No model, no cost, no bias. |5354Full Mode has ONE model judge (Codex), not two. The deterministic gate55is the 3rd judge, pass/fail, never scored. There is no second model56judge in Full Mode — the implementer never scores its own work.5758Do not assume any other provider is available. Check before using one;59if it is unavailable, report it instead of substituting. A third model60may be used only as a final tie-breaker if the user explicitly asks,61and say so in the report.6263Swap implementer and critic on the next run if the user asks, or if one64model stalls. Never let the same model review its own work in the same65round: same family is the same person at a different time of day.66Exception: the explicitly labelled degraded 1-model and no-delegate67modes below, where the same model reviews in a fresh context and the68score is labelled single-judge / self-review, low confidence, never69consensus. That exception does not apply in Full Mode.7071The orchestrator is not a judge in this loop. It drives the agent pane,72reads the result, and decides keep / revert / next step. The user73is the final judge on taste calls.7475## If the Harness Cannot Call the Other Models (Fallback)7677Check availability BEFORE starting the loop, not mid-climb: agy auth,78Codex auth / quota, agent pane reachable, deterministic gate command79exists and runs. Report exactly what is missing. Never invent a80provider, never fake a score, never silently substitute a same-family81model and call it a different judge.8283| Mode | When | What changes |84|---|---|---|85| Full | agy + Codex + deterministic gate all reachable | As in The Loop below. Threshold = model judge >= 9 AND gate passed. |86| Degraded, 1 model | Only one of agy / Codex reachable | Implementer = available model. Critic = same model in a fresh context, different role prompt, read-only. Deterministic gate still hard. Human is final judge. Label every score "single-judge, low confidence". A 9/10 here is NOT consensus — if it hits 9, pause and ask the user whether to accept it as final. Cap at 4 rounds unless the user says more. |87| Degraded, 0 models | No other model reachable, only the orchestrator | No model scores at all. Deterministic gate only. Do not invent scores. Run as `bounded-agent-loop` instead, or stop and ask the user to score it / supply a judge. |88| No delegate tool | Harness cannot spawn subagents / agent pane unavailable | Run sequentially in one pane, one round at a time, with the review in a FRESH context (new session / fresh turn, never the same turn as the implementation). This is single-agent self-review, NOT a second judge: label every score "self-review, low confidence", human is final judge, cap at 4 rounds. If it hits 9, pause and ask the user whether to accept it as final. Threshold here does NOT count as consensus. See `requesting-code-review` fallback. |8990Rules for any degraded mode:9192- Tell the user the mode, which seats are missing, and what that costs93 in confidence — then get their "go" before continuing, unless they94 pre-approved degraded mode for this run.95- Never run at API prices without warning the user first.96- If a provider dies mid-climb: checkpoint the best version, stop the97 round, report what was lost, and ask whether to continue degraded,98 swap implementer / critic, or stop. Do not carry on as if nothing99 happened.100101## If the Goal / Eval Is Unspecified (Walkthrough)102103Do NOT start the loop with a blank goal or eval. Walk the user through104it first. Ask at most 5 questions, one at a time, in plain language.105Suggest a default with every question, so the user can just say "use106that", edit it, or say "I don't know".1071081. What are we improving, in one sentence?1092. What does good look like — what would a 9/10 mean here?1103. How do we check it — tests, a checklist, screenshots, sources?1114. What must NOT change?1125. When do we stop — threshold, rounds, or time?113114If the user is unsure, do the suggesting for them. Concrete defaults115per question — offer these verbatim, do not invent different ones:116117| # | Question | Suggested default |118|---|---|---|119| 1 | What are we improving? | "Improve <the artefact, in the user's words> until the judges score it 9/10" |120| 2 | What does 9/10 mean? | The Model score column from the Define the Eval table for their task type (coding / research / UI / docs) |121| 3 | How do we check it? | The Deterministic gate column from the same table (tests + lint + build; sources checked; screenshots; checklist) |122| 4 | What must NOT change? | Tests, the eval itself, and every file in Immutable — by default, nothing outside Editable files |123| 5 | When do we stop? | Threshold 9/10 from the model judge, gate passed, max 8 rounds, plateau 2 rounds with no gain |124125- Propose a draft goal and eval from the Define the Eval table for126 their task type (coding / research / UI / docs), with defaults:127 threshold 9/10, max 8 rounds, plateau 2.128- If they say "I don't know" to a question, offer 2-3 options,129 recommend one, and say why in one line. Let them pick.130- Fill in the authorization form below with your draft, show it back,131 and get an explicit "go" before starting. The user never has to fill132 a blank form.133- Score the baseline first and show it to them. Seeing a 5/10 and what134 the judges would fix often clarifies the goal faster than more135 questions.136137If after the walkthrough there is still no scoreable eval, stop. Do138not guess one silently. Say what is missing, and offer `bounded-agent-loop`139or a one-pass draft instead.140141## Authorizing a Run142143Needs explicit authorization and hard limits before it starts. Fill every144field — an unfilled budget is not an unlimited budget, it is a reason to145stop and ask.146147```text148Run a hill climb.149150Goal: <one sentence, what "top of the hill" looks like>151Eval: <how to score 0-10, who scores, what counts>152Deterministic gate: <exact command / checklist that must pass first>153Threshold: <default 9/10 from the model judge, gate passed>154Max rounds: <default 8>155Plateau: <default 2 rounds with no gain = stop>156Baseline: <score the starting version — must be scored before the loop starts>157Editable files: <explicit paths the worker may change>158Immutable: <tests, eval, and files it must not touch>159Worktree/branch: <disposable branch or worktree to use>160Cost cap: <limit, or "none enforceable">161Usage pacing: <budget per hour as a share of the ACTIVE provider's162 limit window, e.g. ~20% per hour; never 100% — a run163 that burns the window stalls. The window belongs to164 whichever agent is running (Codex, agy, OpenCode,165 Muse) — agy's is 5h; use the active provider's own166 window. Check remaining budget at every round167 boundary; slow down or pause before the window is168 exhausted>169```170171## Define the Eval172173| Task | Deterministic gate | Model score 0-10 |174|---|---|---|175| Coding | tests + lint + build pass, no regressions | Correctness, clarity, scope discipline, no over-engineering |176| Research idea | Sources checked, claims traceable, gaps listed | Accuracy, evidence, coverage, novelty, falsifiability |177| UI | Screenshots light + dark, desktop + phone, empty + full | Layout, hierarchy, consistency, distinctiveness |178| Docs / writing | Checklist: facts verified, no invented claims | Clarity, completeness, tone, structure |179180Score the baseline first, before any change — in this order: run the181deterministic gate on the baseline, then score it. If the gate fails182on the baseline, the baseline score is reference-only and the loop183does NOT start until the gate passes; fix the baseline first. A184gate-failed baseline score never counts as a kept score.185Every 3rd round, re-score the baseline: if the re-score differs by186more than 1 from the original baseline, report the drift and re-anchor187like this: set baseline = new re-score, set best kept score = new188re-score, and mark all earlier scores as old-scale in the report —189pre-drift and post-drift scores are not comparable. Calibration: models190rate themselves too high. When the user scores the artefact, log the gap191between the model judge's score and the user's score as the calibration192offset, and subtract it from every later model score. Recompute the193offset each time the user scores again; the latest offset replaces the194old one. If a195judge scores something that is not in the screenshots / code / sources196in front of it, discard that finding and say so — check every finding197against the artefact before acting on it.198199## The Loop200201Score rule — read this before step 5. There is ONE model judge and202one deterministic gate, and "the score went up" means all of this:203204- Score = the model judge's 0-10 score for that round. The205 deterministic gate is pass/fail, it is never scored.206- Keep = judge score rose vs the best kept score so far, AND the gate207 passed. Compare only against the last kept round or the baseline —208 never against a reverted round.209- Threshold = model judge scores >= 9 AND the gate passes in the same210 round.211- Plateau = judge score did not rise above the best kept score for 2212 consecutive rounds. Flat is not a gain.213214Checkpoint / revert mechanics. The loop runs ONLY on a disposable215branch / worktree. Before starting, verify the worktree is clean216(`git status --porcelain` empty, except files the loop is about to217edit). Stage ONLY the editable files from the authorization form —218never the whole tree: `git add -- <editable files>`. Checkpoint the219baseline first: commit those files with `git commit -m "hill-climb220baseline"`. Record that commit as the best checkpoint. On every kept221round, commit the editable files again:222`git commit -m "hill-climb round N: <score>/10"`, and update the best223checkpoint. On every revert, go back to the best checkpoint with224`git restore -- <editable files>` plus225`git clean -fd -- <editable files>` (or `git reset --hard226<best-checkpoint>` as a last resort, disposable branch only). These227commands destroy uncommitted state — never run them on a shared228branch, and never run `git add -A` or a bare `git clean -fd` here.229Never let dirty edits from a reverted round contaminate the next230attempt.2312321. Record goal, eval, threshold, max rounds, plateau, baseline score,233 editable files, and immutable eval. Checkpoint the baseline234 (commit on the disposable branch, per above).2352. Implementer (agy first) makes ONE small change toward the judges'236 last notes. Small, or the score change tells you nothing.2373. Run the deterministic gate. Fail = revert (per above), record the238 failure, next attempt. Never weaken the gate, the tests, or the eval239 to get a pass.2404. The model judge scores the new version 0-10, read-only, with241 specific fixes listed. The judge sees only the artefact and the eval242 rubric — no implementation history, no earlier rounds, no notes on243 what changed. A judge that knows what was attempted scores the244 attempt, not the artefact. A fresh review after fixes is a new round.245 Every 3rd round, re-score the baseline (see Define the Eval); if246 drift is more than 1, report it and re-anchor.2475. Keep only if the judge score went up vs the best kept score (see248 Score rule above) AND the gate passed. On keep, commit and update249 the best checkpoint. On revert, reset to the best checkpoint.250 Revert if it went down or flat with no new information. Keep the251 best version seen so far, always.2526. Stop on: threshold reached by the model judge with the gate passed,253 plateau (default 2 rounds with no gain, per Score rule), max rounds,254 budget out, scope change, or missing tools / credentials. Plateau is255 a decision point, not a failure — report the best score and ask the256 user what the top is.257 Theo's line: "once you're climbing the hill, you decide what the top is."2587. Report rounds in a table, best score, what changed, what stalled,259 and the branch / checkpoint. Do not merge or deploy. That is a260 separate authorization.261262## Escaping a Local Peak263264Hill climbing gets stuck. When the score is flat for 2 rounds, the265loop STOPS (plateau). Do not autonomously branch, restructure, or266keep iterating. Present these options to the user and wait for their267pick — only proceed on their explicit pick:268269- Change the approach, not the colour: restructure, split, remove, or270 restart from the best checkpoint with a different strategy.271- Or branch 2-3 different approaches in parallel worktrees, score each272 once, keep the best branch.273- Or stop. A flat score is usually waiting on a human decision, not274 more loops.275276## Scope277278Fight scope creep. The judges will find more things every round; fix279only what moves the eval score for this goal. Park the rest as280follow-ups. A climb that keeps widening never reaches the top.281282## Agent Delegation283284- Work through a visible agent pane, one step at a time, pasted so it285 is visible. Read the result, evaluate it, then paste the next step286 or a correction without asking. Report only when DONE, BLOCKED, or287 needing a decision.288- Never use headless exec (`codex exec`, `opencode run` with a piped289 prompt) for fleet-machine work — use a visible pane. On a standalone290 machine with no agent pane, one headless run is allowed for a291 review / audit step only: prompt from a file, log to the agent's log292 directory, no edits beyond the report. Never hijack another live agent.293- The orchestrator drives; the agent in the pane does the work.294295## Debian / Linux (Standalone) — Addendum296297The shared `skill-authoring` / `skill-create` texts are Windows /298OneDrive-first. Linux / Debian is first-class here too.299300- Resolve the library clone from `~/.config/theprawnskills/library.json`301 (`library` key), never guess the path.302- Installed skills: `~/.agents/skills` (Linux, symlinks). Never edit303 an installed copy; edit the clone.304- Installer (Python 3.11+): run the repository installer from the library305 root — see `platforms/linux/README.md` for the exact306 `--dry-run` / `--apply` / `--check` commands. Auto mode is symlinks on307 Linux, copies on Windows.308- Shell: bash with `python3` on Linux, PowerShell on Windows.309 Do not copy Windows paths, OneDrive paths, or `dotagents sync`310 into a Linux run. Do not use the Windows machine registry311 for standalone propagation.312- This skill is ON-DEMAND, not in `default-profile.toml`, unless313 daily exposure is requested. Keep `default-profile.toml` as the314 shared daily selection; check all supported OSes before changing315 profile membership.316317## Related Skills318319- `bounded-agent-loop` — the closest existing skill; bounded repair320 loop with an acceptance command, max 3 attempts. Use it when the321 eval is pass/fail, not a score.322- `skill-judge` — multi-dimensional scoring; the scoring half of a323 council, for skills.324- `requesting-code-review` — review, fix, re-review loop.325- `skill-create`, `skill-authoring` — how this skill was made.326 Standalone variants live under `platforms/linux/skills/`.327- `skill-creator` — generic upstream skill-creation loop328 (draft, test, eval, rewrite, repeat). Method only; its Claude-only329 eval tooling and paths do not apply here.330- `skill-add` — import an existing skill from GitHub / a link / pasted331 text. Not used here: this skill is new, not imported, and Theo's332 `council` was adapted, not copied verbatim.333334## Provenance335336Adapted from Theo Browne's `council` skill (t3dotgg/tstack, MIT) and337the video "Agents Sucked At UI (Until Today)", 2026-11-03. Generalised338from UI-only to any scoreable goal, and from three providers to one339implementer provider, one model-judge provider, and one deterministic340gate.