Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions .agents/skills/arena/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,7 @@ Every candidate gets the same prompt, so the prompt is the contract.

1. State the artifact each candidate is producing.
2. Derive the rubric, 3-6 concrete gradeable criteria for what success looks like on *this* task. It is the picker's tool in Phase D. Candidates only see the task.
3. Pick the runners from the `arena runners` line in `.agents/pstack-models.md`, then `~/.agents/pstack-models.md`, by default `claude-fable-5-2-high`, `claude-opus-5-5-high-fast`, and `gemini-4-carbon-high`. An `auto` or `inherit-parent` entry is the parent model, so omit `model` and spawn `generalPurpose`. If the Task tool rejects a slug, use the nearest valid slug of the same family or drop the seat, keep at least two runners, and say so. Families go by prefix, and the longest match wins: `claude-fable-*`, `claude-*`, and `gemini-*`, after dropping a leading `cursor-`. Any other slug's family is its text before the first hyphen. Spawn more for multiple design directions, and repeat one model when the work is generation-bound, not judgment-sensitive. Pick each subagent's reasoning level per the Subagents section of the **poteto-mode** skill.
3. Pick the runners from the `arena runners` line in `.agents/pstack-models.md`, then `~/.agents/pstack-models.md`, by default `claude-fable-5-2-high`, `claude-opus-5-5-high-fast`, and `gemini-4-neon-xhigh`. An `auto` or `inherit-parent` entry is the parent model, so omit `model` and spawn `generalPurpose`. If the Task tool rejects a slug, use the nearest valid slug of the same family or drop the seat, keep at least two runners, and say so. Families go by prefix, and the longest match wins: `claude-fable-*`, `claude-*`, and `gemini-*`, after dropping a leading `cursor-`. Any other slug's family is its text before the first hyphen. Spawn more for multiple design directions, and repeat one model when the work is generation-bound, not judgment-sensitive. Pick each subagent's reasoning level per the Subagents section of the **poteto-mode** skill.
4. Give each candidate its own output location (a git worktree where possible, otherwise `/tmp/arena-<slug>/candidate-<n>/`), per the **separate-before-serializing-shared-state** principle skill.

## Phase B: Fan out
Expand All @@ -26,7 +26,7 @@ If a candidate produces no output, proceed with N-1 and note the dropout.

## Phase C: Cross-judge

After all Phase B candidates complete, spawn one readonly judge from the `arena cross-judge pool` line in `.agents/pstack-models.md`, then `~/.agents/pstack-models.md`, by default `claude-opus-5-5-high-fast`, `gemini-4-carbon-high`. The judge is never from the parent's vendor, and Fable counts as Claude for that check, so use the first entry from another vendor. An `auto` or `inherit-parent` entry is the parent model. If you do not know the parent's model, use the first real slug and say that the vendor check did not run. If the Task tool rejects a slug, use the nearest valid slug of the same family, then the next such entry, or skip the judge, and say so. While you read in Phase D, the judge scores each candidate, by path label, on each rubric criterion and recommends a base.
After all Phase B candidates complete, spawn one readonly judge from the `arena cross-judge pool` line in `.agents/pstack-models.md`, then `~/.agents/pstack-models.md`, by default `claude-opus-5-5-high-fast`, `gemini-4-neon-xhigh`. The judge is never from the parent's vendor, and Fable counts as Claude for that check, so use the first entry from another vendor. An `auto` or `inherit-parent` entry is the parent model. If you do not know the parent's model, use the first real slug and say that the vendor check did not run. If the Task tool rejects a slug, use the nearest valid slug of the same family, then the next such entry, or skip the judge, and say so. While you read in Phase D, the judge scores each candidate, by path label, on each rubric criterion and recommends a base.

## Phase D: Pick a base

Expand Down
4 changes: 2 additions & 2 deletions .agents/skills/poteto-mode/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -90,11 +90,11 @@ Read the leaf skill in full for any principle you apply. Each entry names when i

- Code, prose, and judgment: `claude-opus-5-5-xhigh-fast`.
- Trivial mechanical edits: `claude-opus-5-5-high-fast`.
- Diverse review and design panels: one runner each on `claude-fable-5-2-high`, `claude-opus-5-5-high-fast`, and `gemini-4-carbon-high`.
- Diverse review and design panels: one runner each on `claude-fable-5-2-high`, `claude-opus-5-5-high-fast`, and `gemini-4-neon-xhigh`.

**Fable 5.2 is an exception.** It runs outside a panel only when the user named it for this task or two Opus 5.5 xhigh runs of the same brief failed its check. Say which. Difficulty alone never qualifies.

**Match the model to the job.** Start from the role's line or default, keep its family, and set the level the job needs. The `/setup-pstack` budget sets each role's starting level. Panel seats keep their listed level. Raise any other role one step, up to xhigh, for the hardest or subtlest work, lower it one step for judgment, and two or more steps for routine work. No role runs above xhigh. Opus 5.5 always runs its `-fast` slug. Verification always runs on the `swarm workers` model, `claude-opus-5-5-high-fast` by default. A Gemini seat never runs above `gemini-4-carbon-high`, its top slug. If the Task tool rejects a slug, never pass it again. Fall back to the role's line or default, then the calling skill's fallback, then the closest valid slug of the same family, then omit `model`. Say so.
**Match the model to the job.** Start from the role's line or default, keep its family, and set the level the job needs. The `/setup-pstack` budget sets each role's starting level. Panel seats keep their listed level. Raise any other role one step, up to xhigh, for the hardest or subtlest work, lower it one step for judgment, and two or more steps for routine work. No role runs above xhigh. Opus 5.5 always runs its `-fast` slug. Verification always runs on the `swarm workers` model, `claude-opus-5-5-high-fast` by default. A Gemini seat runs at `gemini-4-neon-xhigh` unless its line names another level. If the Task tool rejects a slug, never pass it again. Fall back to the role's line or default, then the calling skill's fallback, then the closest valid slug of the same family, then omit `model`. Say so.

You own every subagent's work. Review the diff and write your own summary. A second opinion is the same prompt on another vendor's model.

Expand Down
4 changes: 3 additions & 1 deletion .agents/skills/setup-pstack/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,8 @@ Use the actual subagent tool schema or a model-list capability exposed by this C

Read project configuration first, otherwise personal configuration. The default role-to-model mapping is the shape in section 3. If a configuration file already exists, read it and treat its `# budget` line and its role values as the current choices. Otherwise start from those defaults. A line whose role is not in section 3, such as `hardest tasks`, `architect runners`, `reflect tooling`, or `why investigators`, is from a retired role. Drop it.

Older defaults put another Gemini model, then `gemini-4-neon-high`, in the panel seat. So in any rule, a `gemini-4-neon-high` entry, or a `gemini-4-*` entry other than `gemini-4-neon-*`, in a panel role (`arena runners`, `arena cross-judge pool`, `interrogate reviewers`) is an old default. Each one becomes `gemini-4-neon-xhigh`. A Gemini 4 Neon entry at any other level is the user's choice and stays. Tell the user which entries you replaced.

A file without the `# defaults: 3` line comes from an older setup. That setup wrote a line for every role, so some of its values are old defaults, not choices. Before section 2, replace these values:

- An `arena cross-judge pool` list of exactly one `claude-fable-*` entry, one `claude-opus-*` entry, and one `gemini-*` entry, in any order, takes the role's section 3 list. Fable and Opus are one vendor, so the Fable entry could judge only where the Opus entry also could.
Expand All @@ -35,7 +37,7 @@ Use the available user-input tool, or a concise text question when it is unavail
- `medium` — high reasoning when supported
- `small` — medium reasoning when supported

A budget changes only explicit reasoning levels. `inherit-parent` and `auto` do not change. For an explicit entry, move its reasoning level on the ladder `xhigh` > `high` > `medium` > `low`: `large` keeps it, `medium` lowers it one step, and `small` lowers it two steps. No role runs above `xhigh`, and a target at the bottom of the ladder stays there. Choose a supported level at or below the target. If none is known, leave effort inherited and disclose it. Model IDs and reasoning are separate values; never construct an effort-suffixed model slug.
A budget changes only explicit reasoning levels. `inherit-parent` and `auto` do not change. For an explicit entry, move its reasoning level on the ladder `xhigh` > `high` > `medium` > `low`: `large` keeps it, `medium` lowers it one step, and `small` lowers it two steps. No role runs above `xhigh`, and a target at the bottom of the ladder stays there. Choose a supported level at or below the target. If none is known, leave effort inherited and disclose it. Model IDs and reasoning are separate values; never construct an effort-suffixed model slug. Gemini 4 Neon slugs run from `low` to `xhigh` with no `max` or `fast` slug, so its seat runs at `gemini-4-neon-xhigh`, its top slug. A Gemini 4 Neon entry in a panel role takes its own target, `xhigh` moved by the budget, so `large` keeps `gemini-4-neon-xhigh`, `medium` writes `gemini-4-neon-high`, and `small` writes `gemini-4-neon-medium`.

Show all roles and their values, plus any removed retired roles. Let the user accept or change specific roles. A scalar role takes `inherit-parent`, `auto`, `<validated-model>`, or `<validated-model> <validated-effort>`. Panel roles take comma-separated entries; every entry, including aliases, creates one seat. `arena cross-judge pool` is a pool, but Arena runs one judge. It tries only values from a vendor other than the parent's, in list order, so order it by preference. Fable and the other Claude models are one vendor. An `inherit-parent` or `auto` value never judges.

Expand Down
2 changes: 1 addition & 1 deletion .agents/skills/statechart/SKILL.md
Original file line number Diff line number Diff line change
Expand Up @@ -17,7 +17,7 @@ Work in a scratch folder outside the repo. Write a TLA+ spec and its config. The
<skill dir>/scripts/statechart SPEC.tla [--config FILE] [--view FILE] [--out DIR]
```

`scripts/statechart.ts` is TypeScript, so run it with a runtime that can execute TypeScript.
`scripts/statechart.ts` is TypeScript, so run it with a runtime that can execute TypeScript. `scripts/statechart` is a POSIX shell script that picks one. On native Windows, run `node --disable-warning=ExperimentalWarning <skill dir>/scripts/statechart.ts` with Node 22.18 to 22.x, or Node 23.6 or later.

The first run installs TLC and Java with mise and needs the network.

Expand Down
5 changes: 4 additions & 1 deletion .agents/skills/statechart/scripts/frontends/dot.ts
Original file line number Diff line number Diff line change
Expand Up @@ -35,7 +35,10 @@ export const parseState = (label: string, parsed = new Map<string, [string, Valu

export type Lines = string | Iterable<string>;

export const linesOf = (input: Lines) => (typeof input === "string" ? input.split("\n") : input);
export const splitLines = (text: string) =>
(text.endsWith("\r") ? text.slice(0, -1) : text).split(/\r?\n/);

export const linesOf = (input: Lines) => (typeof input === "string" ? splitLines(input) : input);

export function parseDot(input: Lines): Graph {
const states = new Map<string, State>();
Expand Down
13 changes: 13 additions & 0 deletions .agents/skills/statechart/scripts/janitor.ts
Original file line number Diff line number Diff line change
@@ -0,0 +1,13 @@
import fs from "node:fs";
import { finished } from "node:stream/promises";
import { killGroup } from "./kill.ts";

const [pid, ...dirs] = process.argv.slice(2);

let input = "";
process.stdin.setEncoding("utf8").on("data", (chunk: string) => (input += chunk));
await finished(process.stdin).catch(() => {});
const running = !input.includes("exited");
const groupOutlivesWorker = process.platform !== "win32";
if (running || groupOutlivesWorker) await killGroup(Number(pid));
for (const dir of dirs) fs.rmSync(dir, { recursive: true, force: true });
15 changes: 15 additions & 0 deletions .agents/skills/statechart/scripts/kill.ts
Original file line number Diff line number Diff line change
@@ -0,0 +1,15 @@
import { spawn } from "node:child_process";

export const killGroup = (pid: number) =>
new Promise<void>((resolve) => {
if (process.platform === "win32") {
spawn("taskkill", ["/pid", String(pid), "/t", "/f"], { stdio: "ignore" })
.on("error", () => resolve())
.on("exit", () => resolve());
return;
}
try {
process.kill(-pid, "SIGKILL");
} catch {}
resolve();
});
24 changes: 9 additions & 15 deletions .agents/skills/statechart/scripts/run.ts
Original file line number Diff line number Diff line change
Expand Up @@ -8,7 +8,7 @@ import { inspect } from "node:util";
import { boundRun, chartRun, configText } from "./backends/cfg.ts";
import { plainText } from "./backends/plain-text.ts";
import { readConfig } from "./frontends/config.ts";
import { parseDot } from "./frontends/dot.ts";
import { parseDot, splitLines } from "./frontends/dot.ts";
import { readCounterexample, readMarks } from "./frontends/marks.ts";
import { readMessages } from "./frontends/messages.ts";
import { initialOnlyProperties, readChecks, readSany, readSpec } from "./frontends/spec.ts";
Expand Down Expand Up @@ -55,10 +55,10 @@ export function* fileLines(file: string) {
kept = end;
continue;
}
yield* buf.toString("utf8", 0, cut).split("\n");
yield* splitLines(buf.toString("utf8", 0, cut));
kept = buf.copy(buf, 0, cut + 1, end);
}
yield buf.toString("utf8", 0, kept);
yield* splitLines(buf.toString("utf8", 0, kept));
} finally {
fs.closeSync(fd);
}
Expand All @@ -82,7 +82,7 @@ const runChild = (file: string, args: string[], options: { cwd: string; env: Nod
execFile(
file,
args,
{ encoding: "utf8", maxBuffer: 1 << 30, ...options },
{ encoding: "utf8", maxBuffer: 1 << 30, windowsHide: true, ...options },
(error, stdout, stderr) =>
resolve({
status: error ? (typeof error.code === "number" ? error.code : null) : 0,
Expand All @@ -93,14 +93,6 @@ const runChild = (file: string, args: string[], options: { cwd: string; env: Nod
);
});

export const killGroup = (pid: number) =>
new Promise((resolve) =>
spawn("/bin/sh", ["-c", 'kill -s KILL -- "-$1"', "sh", String(pid)], { stdio: "ignore" }).on(
"exit",
resolve,
),
);

export const messageOf = (error: unknown) =>
typeof error === "object" &&
error !== null &&
Expand Down Expand Up @@ -142,10 +134,10 @@ function* outputLines(output: Output) {
rest = buf;
continue;
}
yield* utf8.decode(buf.subarray(0, cut)).split("\n");
yield* splitLines(utf8.decode(buf.subarray(0, cut)));
rest = buf.subarray(cut + 1);
}
yield utf8.decode(rest);
yield* splitLines(utf8.decode(rest));
}

const messagesOf = (output: Output): Iterable<Part> => ({
Expand All @@ -170,7 +162,7 @@ const runToFiles = (
for (const f of files) fds.push(fs.openSync(f, "w"));
let child: ChildProcess;
try {
child = spawn(file, args, { ...options, stdio: ["ignore", ...fds] });
child = spawn(file, args, { ...options, stdio: ["ignore", ...fds], windowsHide: true });
} finally {
for (const fd of fds) fs.closeSync(fd);
}
Expand Down Expand Up @@ -298,6 +290,8 @@ export async function check(specFile: string, configFile: string): Promise<Check
[
`-Djava.io.tmpdir=${tmp}`,
`-DTLA-Library=${folder}`,
"-Dstdout.encoding=UTF-8",
"-Dstderr.encoding=UTF-8",
...jvm,
"-cp",
tlc.jar,
Expand Down
27 changes: 14 additions & 13 deletions .agents/skills/statechart/scripts/statechart.ts
Original file line number Diff line number Diff line change
Expand Up @@ -4,7 +4,8 @@ import fs from "node:fs";
import os from "node:os";
import path from "node:path";
import { parseArgs } from "node:util";
import { killGroup, messageOf } from "./run.ts";
import { killGroup } from "./kill.ts";
import { messageOf } from "./run.ts";

const USAGE = `Usage: statechart SPEC.tla [--config FILE] [--view FILE] [--out DIR]
Expand Down Expand Up @@ -110,26 +111,26 @@ try {
{
detached: true,
stdio: ["ignore", "inherit", "inherit"],
env: { ...process.env, TMPDIR: tmp },
env: { ...process.env, TMPDIR: tmp, TEMP: tmp, TMP: tmp },
},
);
janitor = spawn(
"/bin/sh",
[
"-c",
'while read -r line; do :; done; kill -s KILL -- "-$1"; rm -rf -- "$2" "$3"',
"sh",
String(worker.pid),
stage,
tmp,
],
process.execPath,
[...runtime, path.join(import.meta.dirname, "janitor.ts"), String(worker.pid), stage, tmp],
{ detached: true, stdio: ["pipe", "ignore", "ignore"] },
);
const toJanitor = janitor.stdin;
toJanitor?.on("error", () => {});
const exited = new Promise<number>((resolve) =>
worker.on("exit", (code) => resolve(code ?? 2)),
worker.on("exit", (code) => {
toJanitor?.write("exited\n");
resolve(code ?? 2);
}),
);
await Promise.race([stopped, exited]);
await killGroup(worker.pid as number);
const running = worker.exitCode === null && worker.signalCode === null;
const groupOutlivesWorker = process.platform !== "win32";
if (worker.pid !== undefined && (running || groupOutlivesWorker)) await killGroup(worker.pid);
const verdict = await exited;
if (stop) {
console.error(`${stop} stopped the check, so the command wrote nothing.`);
Expand Down
29 changes: 29 additions & 0 deletions .agents/skills/statechart/scripts/test/cli.test.ts
Original file line number Diff line number Diff line change
Expand Up @@ -102,6 +102,35 @@ for (const name of fs.readdirSync(corpus)) {
});
}

function asTlcPrintsOnWindows(text: string) {
const lines = text.split("\n");
let depth = 0;
for (let i = 0; i < lines.length - 1; i++) {
if (/@!@!@STARTMSG \d+:\d+ @!@!@$/.test(lines[i])) depth++;
else if (/^@!@!@ENDMSG \d+ @!@!@$/.test(lines[i])) depth--;
if (!depth) lines[i] += "\r";
}
return lines.join("\n");
}

for (const name of fs.readdirSync(corpus)) {
test(`${name} writes the committed chart and report from TLC's output with Windows line ends`, () => {
const dir = copyCase(name, (d) => {
for (const f of fs.readdirSync(d))
if (f.endsWith(".out"))
fs.writeFileSync(path.join(d, f), asTlcPrintsOnWindows(read(path.join(d, f))));
});
const r = run(dir);
assert.equal(r.stderr, "");
const first = path.join(dir, "verdict-1.exit");
const plain = fs.existsSync(first) ? Number(read(first)) : 0;
assert.equal(r.status, plain === 0 ? 0 : [11, 12, 13].includes(plain) ? 1 : 2);
assert.equal(r.report, read(path.join(corpus, name, "report.md")));
assert.equal(read(path.join(r.out, "chart.mmd")), read(path.join(corpus, name, "chart.mmd")));
for (const f of r.files) assert.doesNotMatch(read(path.join(r.out, f)), /\r|@!@!@/, f);
});
}

const launcher = path.join(here, "../statechart");

const launcherTools = () => {
Expand Down
14 changes: 14 additions & 0 deletions .agents/skills/statechart/scripts/test/messages.test.ts
Original file line number Diff line number Diff line change
@@ -1,13 +1,27 @@
import assert from "node:assert/strict";
import fs from "node:fs";
import path from "node:path";
import { test } from "node:test";
import { fileURLToPath } from "node:url";
import { plainText, runtime } from "../backends/plain-text.ts";
import { readMessages } from "../frontends/messages.ts";
import { readVerdict } from "../frontends/verdict.ts";

const here = path.dirname(fileURLToPath(import.meta.url));

const tool = (code: number, level: number, ...lines: string[]) =>
[`@!@!@STARTMSG ${code}:${level} @!@!@`, ...lines, `@!@!@ENDMSG ${code} @!@!@`].join("\n");

const plain = (text: string) => [...plainText(readMessages(text))].join("");

test("TLC output with Windows line ends names the rule that breaks and prints as with Unix ones", () => {
const unix = fs.readFileSync(path.join(here, "corpus", "FirstRule", "verdict-1.out"), "utf8");
const windows = unix.replaceAll("\n", "\r\n");
assert.deepEqual(readVerdict(readMessages(windows)).broke, { kind: "invariant", name: "NotOne" });
assert.deepEqual(readVerdict(readMessages(windows)), readVerdict(readMessages(unix)));
assert.equal(plain(windows), plain(unix));
});

test("TLC's -tool output prints as plain TLC prints it", () => {
const text = [
tool(2262, 0, "TLC2 Version 2.19 of 08 August 2024 (rev: 5a47802)"),
Expand Down
2 changes: 1 addition & 1 deletion .codex-plugin/plugin.json
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
{
"name": "pstack",
"version": "0.15.43",
"version": "0.15.46",
"description": "if you want to go fast, go deep first. pstack helps you write less, but higher quality code. rigorous agent workflows you can parallelize with confidence.",
"author": {
"name": "Lauren Tan"
Expand Down
Loading
Loading