humanish 0.105.0 → 0.106.0

package/dist/listen.js

@@ -1,84 +0,0 @@
-// One loopback bind for every humanish server, with the one failure people actually hit named.
-//
-// `listen EADDRINUSE` used to surface as HUMANISH_UNEXPECTED, the command-boundary catch-all for
-// "a handler threw" (#484). Something already serving on the port is the most expected condition
-// a serve command has: a second session, a forgotten background one, someone's dev server. It only
-// bites with an explicit --port, which is exactly when a person is asking for a stable address.
-export class PortInUseError extends Error {
-    port;
-    holder;
-    constructor(port, holder) {
-        super(holder === "humanish"
-            ? `port ${port} is already served by another humanish process on this machine; stop it, or pass --port 0 for a free port`
-            : `something else is already listening on 127.0.0.1:${port}; pass --port 0 for a free port, or stop it`);
-        this.port = port;
-        this.holder = holder;
-        this.name = "PortInUseError";
-    }
-}
-/**
- * Is the process on this port one of ours? humanish's serve and observer servers answer
- * `/_humanish/history.json` (serve) or `/observer-data.json` (observer) as JSON with a humanish
- * schema; anything else is somebody's app. Bounded to half a second and never throws: a probe
- * that cannot decide says "other", which is the answer that suggests the safer action.
- */
-export async function probePortHolder(port, fetchFn = fetch) {
-    // The whole probe is raced against a hard clock as well: a fetch that ignores its abort signal
-    // (a test double, a proxy) must not turn "tell me whose port this is" into a hang.
-    let hardTimer;
-    const hardStop = new Promise((resolve) => {
-        hardTimer = setTimeout(() => resolve("other"), 1_500);
-        hardTimer.unref?.();
-    });
-    try {
-        return await Promise.race([probeRoutes(port, fetchFn), hardStop]);
-    }
-    finally {
-        clearTimeout(hardTimer);
-    }
-}
-async function probeRoutes(port, fetchFn) {
-    for (const route of ["/_humanish/history.json", "/observer-data.json"]) {
-        const controller = new AbortController();
-        const timer = setTimeout(() => controller.abort(), 500);
-        try {
-            const response = await fetchFn(`http://127.0.0.1:${port}${route}`, { signal: controller.signal });
-            if (!response.ok)
-                continue;
-            const body = (await response.json());
-            if (typeof body?.schema === "string" && body.schema.startsWith("humanish."))
-                return "humanish";
-        }
-        catch {
-            // not ours, or not answering: keep probing, then say "other"
-        }
-        finally {
-            clearTimeout(timer);
-        }
-    }
-    return "other";
-}
-/** Bind on loopback only; resolve the port that was actually bound; name a taken port. */
-export function listenOnLoopback(server, port, probe = probePortHolder) {
-    return new Promise((resolve, reject) => {
-        const onError = (error) => {
-            if (error.code === "EADDRINUSE") {
-                void probe(port).then((holder) => reject(new PortInUseError(port, holder)));
-                return;
-            }
-            reject(error);
-        };
-        server.once("error", onError);
-        // Never beyond loopback; exposure only ever happens through a tunnel or proxy forwarding here.
-        server.listen(port, "127.0.0.1", () => {
-            server.off("error", onError);
-            const address = server.address();
-            if (!address || typeof address === "string") {
-                reject(new Error("server did not bind to a TCP port"));
-                return;
-            }
-            resolve(address.port);
-        });
-    });
-}
-//# sourceMappingURL=listen.js.map

package/dist/listen.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"listen.js","sourceRoot":"","sources":["../src/listen.ts"],"names":[],"mappings":"AAAA,+FAA+F;AAC/F,EAAE;AACF,iGAAiG;AACjG,iGAAiG;AACjG,mGAAmG;AACnG,gGAAgG;AAMhG,MAAM,OAAO,cAAe,SAAQ,KAAK;IAClB;IAAuB;IAA5C,YAAqB,IAAY,EAAW,MAAkB;QAC5D,KAAK,CACH,MAAM,KAAK,UAAU;YACnB,CAAC,CAAC,QAAQ,IAAI,2GAA2G;YACzH,CAAC,CAAC,oDAAoD,IAAI,6CAA6C,CAC1G,CAAC;QALiB,SAAI,GAAJ,IAAI,CAAQ;QAAW,WAAM,GAAN,MAAM,CAAY;QAM5D,IAAI,CAAC,IAAI,GAAG,gBAAgB,CAAC;IAC/B,CAAC;CACF;AAED;;;;;GAKG;AACH,MAAM,CAAC,KAAK,UAAU,eAAe,CAAC,IAAY,EAAE,UAAwB,KAAK;IAC/E,+FAA+F;IAC/F,mFAAmF;IACnF,IAAI,SAAoD,CAAC;IACzD,MAAM,QAAQ,GAAG,IAAI,OAAO,CAAa,CAAC,OAAO,EAAE,EAAE;QACnD,SAAS,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,KAAK,CAAC,CAAC;QACtD,SAAS,CAAC,KAAK,EAAE,EAAE,CAAC;IACtB,CAAC,CAAC,CAAC;IACH,IAAI,CAAC;QACH,OAAO,MAAM,OAAO,CAAC,IAAI,CAAC,CAAC,WAAW,CAAC,IAAI,EAAE,OAAO,CAAC,EAAE,QAAQ,CAAC,CAAC,CAAC;IACpE,CAAC;YAAS,CAAC;QACT,YAAY,CAAC,SAAS,CAAC,CAAC;IAC1B,CAAC;AACH,CAAC;AAED,KAAK,UAAU,WAAW,CAAC,IAAY,EAAE,OAAqB;IAC5D,KAAK,MAAM,KAAK,IAAI,CAAC,yBAAyB,EAAE,qBAAqB,CAAC,EAAE,CAAC;QACvE,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;QACzC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,GAAG,CAAC,CAAC;QACxD,IAAI,CAAC;YACH,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,oBAAoB,IAAI,GAAG,KAAK,EAAE,EAAE,EAAE,MAAM,EAAE,UAAU,CAAC,MAAM,EAAE,CAAC,CAAC;YAClG,IAAI,CAAC,QAAQ,CAAC,EAAE;gBAAE,SAAS;YAC3B,MAAM,IAAI,GAAG,CAAC,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAyB,CAAC;YAC7D,IAAI,OAAO,IAAI,EAAE,MAAM,KAAK,QAAQ,IAAI,IAAI,CAAC,MAAM,CAAC,UAAU,CAAC,WAAW,CAAC;gBAAE,OAAO,UAAU,CAAC;QACjG,CAAC;QAAC,MAAM,CAAC;YACP,6DAA6D;QAC/D,CAAC;gBAAS,CAAC;YACT,YAAY,CAAC,KAAK,CAAC,CAAC;QACtB,CAAC;IACH,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,0FAA0F;AAC1F,MAAM,UAAU,gBAAgB,CAC9B,MAAc,EACd,IAAY,EACZ,QAA+C,eAAe;IAE9D,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACrC,MAAM,OAAO,GAAG,CAAC,KAA4B,EAAQ,EAAE;YACrD,IAAI,KAAK,CAAC,IAAI,KAAK,YAAY,EAAE,CAAC;gBAChC,KAAK,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,cAAc,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC;gBAC5E,OAAO;YACT,CAAC;YACD,MAAM,CAAC,KAAK,CAAC,CAAC;QAChB,CAAC,CAAC;QACF,MAAM,CAAC,IAAI,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;QAC9B,+FAA+F;QAC/F,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,WAAW,EAAE,GAAG,EAAE;YACpC,MAAM,CAAC,GAAG,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;YAC7B,MAAM,OAAO,GAAG,MAAM,CAAC,OAAO,EAAE,CAAC;YACjC,IAAI,CAAC,OAAO,IAAI,OAAO,OAAO,KAAK,QAAQ,EAAE,CAAC;gBAC5C,MAAM,CAAC,IAAI,KAAK,CAAC,mCAAmC,CAAC,CAAC,CAAC;gBACvD,OAAO;YACT,CAAC;YACD,OAAO,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;QACxB,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;AACL,CAAC"}

package/dist/local-agent-claude-session.d.ts

@@ -1,45 +0,0 @@
-import { spawn, type ChildProcessWithoutNullStreams } from "node:child_process";
-import type { ActorCapabilities } from "./actor-contract.js";
-import type { CuaProvider, CuaTurn } from "./computer-use.js";
-import type { ReasoningEffort } from "./reasoning-effort.js";
-type JsonObject = Record<string, unknown>;
-/** The transport, injected so tests drive a fake session with no CLI and no spend. */
-export interface ClaudeStreamTransport {
-    /** Write one NDJSON message to the session. */
-    send(message: JsonObject): void;
-    /** Resolves with the next `result` message; rejects if the session ends first or the clock runs out. */
-    awaitResult(timeoutMs: number, signal?: AbortSignal): Promise<JsonObject>;
-    close(): void;
-}
-/** NDJSON over the child's stdio. Everything that is not a `result` is a progress line and is ignored. */
-export declare function stdioClaudeTransport(child: ChildProcessWithoutNullStreams): ClaudeStreamTransport;
-export interface ClaudeSessionOptions {
-    /** Absent = spawn `claude`. Injected in tests. */
-    transport?: ClaudeStreamTransport;
-    model?: string;
-    /** Recorded on the trace; Claude Code's `-p` mode takes no effort flag, so this is what was asked for, not applied. */
-    reasoningEffort?: ReasoningEffort;
-    /** Per-turn wall clock. */
-    timeoutMs?: number;
-    /** Scratch root for the session's working directory (the screenshots it is allowed to Read). */
-    workRoot?: string;
-    /** Extra arguments, e.g. a model override. Tests never pass any. */
-    spawnFn?: typeof spawn;
-}
-export declare const CLAUDE_SESSION_CAPABILITIES: ActorCapabilities;
-export interface ClaudeSession {
-    provider: CuaProvider;
-    /** Ends the process and removes its scratch directory. The run owns the lifetime, not the provider. */
-    close(): Promise<void>;
-}
-/** The message shape Claude Code reads on stdin in stream-json mode. */
-export declare function userMessage(text: string): JsonObject;
-/**
- * Start one Claude Code session and return a provider that spends it, one user message per
- * computer-use turn. Started EAGERLY (before the first screenshot) so the CLI's own boot is paid
- * while the sandbox is still settling rather than inside turn one.
- */
-export declare function startClaudeSession(options?: ClaudeSessionOptions): Promise<ClaudeSession>;
-/** Read a stream-json `result` message into a CuaTurn. Exported for tests. */
-export declare function turnFromResult(result: JsonObject): CuaTurn;
-export {};

package/dist/local-agent-claude-session.js

@@ -1,203 +0,0 @@
-// ONE Claude Code process as the computer-use brain, instead of a fresh `claude -p` per turn.
-//
-// WHY (#520): `actors[].localAgent: codex` already runs through a persistent app-server thread,
-// so the participant remembers what it tried. `claude` spawned `claude -p` per turn, so EVERY
-// turn started cold. Measured on the same lab with the same credentials (n=1 each): one-shot,
-// 188 actions over 90 turns and never finished; a thread that remembers, 21 actions over 8 turns
-// and goal_satisfied in 103 s. A participant that cannot remember trying the menu tries the menu
-// again. Until this existed, comparing the two agents measured the transport, not the model.
-//
-// THE MECHANISM, checked on this machine before it was written: `claude -p --input-format
-// stream-json --output-format stream-json --verbose` is a bidirectional session over stdio. One
-// NDJSON `user` message in, a stream of `system` / `assistant` / `result` messages out, then it
-// waits for the next `user` message with the conversation intact. Two messages, one session id,
-// and a codeword given in the first turn was recalled in the second. `--allowedTools Read` keeps
-// the bound the one-shot version had: it looks at a picture, and that is all it can do.
-//
-// What this is NOT: a change to the loop, the executor, the trace, or the Observer. Only where
-// the next action comes from.
-import { spawn } from "node:child_process";
-import { mkdtemp, rm, unlink, writeFile } from "node:fs/promises";
-import { tmpdir } from "node:os";
-import path from "node:path";
-import readline from "node:readline";
-import { declaredOutcomeOf, parseAgentJson, promptFor, toCuaActions, LOCAL_AGENT_CAPABILITIES } from "./local-agent-cli.js";
-/** NDJSON over the child's stdio. Everything that is not a `result` is a progress line and is ignored. */
-export function stdioClaudeTransport(child) {
-    const rl = readline.createInterface({ input: child.stdout });
-    let waiter;
-    let stderrTail = "";
-    let exited;
-    const fail = (error) => {
-        const pending = waiter;
-        waiter = undefined;
-        pending?.reject(error);
-    };
-    rl.on("line", (line) => {
-        let message;
-        try {
-            message = JSON.parse(line);
-        }
-        catch {
-            return; // never a reason to end a run
-        }
-        if (message.type !== "result")
-            return;
-        const pending = waiter;
-        waiter = undefined;
-        pending?.resolve(message);
-    });
-    child.stderr.on("data", (chunk) => {
-        stderrTail = (stderrTail + chunk.toString("utf8")).slice(-400);
-    });
-    child.on("close", (code, signal) => {
-        exited = `Claude Code exited ${code ?? `on ${signal ?? "a signal"}`}${stderrTail.trim() ? `: ${stderrTail.trim()}` : ""}`;
-        // Fail loud with the CLI's own words: a rate-limited plan says so here, which is a sentence
-        // the operator can act on, unlike "turn failed".
-        fail(new Error(exited));
-    });
-    return {
-        send(message) {
-            if (exited !== undefined)
-                throw new Error(exited);
-            child.stdin.write(`${JSON.stringify(message)}\n`);
-        },
-        awaitResult(timeoutMs, signal) {
-            if (exited !== undefined)
-                return Promise.reject(new Error(exited));
-            return new Promise((resolve, reject) => {
-                const timer = setTimeout(() => fail(new Error(`Claude Code produced no result within ${timeoutMs}ms`)), timeoutMs);
-                const onAbort = () => fail(new Error("run stopped"));
-                signal?.addEventListener("abort", onAbort, { once: true });
-                waiter = {
-                    resolve: (value) => { clearTimeout(timer); signal?.removeEventListener("abort", onAbort); resolve(value); },
-                    reject: (error) => { clearTimeout(timer); signal?.removeEventListener("abort", onAbort); reject(error); }
-                };
-            });
-        },
-        close() {
-            rl.close();
-            child.stdin.end();
-            child.kill();
-        }
-    };
-}
-export const CLAUDE_SESSION_CAPABILITIES = LOCAL_AGENT_CAPABILITIES;
-/** The message shape Claude Code reads on stdin in stream-json mode. */
-export function userMessage(text) {
-    return { type: "user", message: { role: "user", content: [{ type: "text", text }] } };
-}
-/**
- * Start one Claude Code session and return a provider that spends it, one user message per
- * computer-use turn. Started EAGERLY (before the first screenshot) so the CLI's own boot is paid
- * while the sandbox is still settling rather than inside turn one.
- */
-export async function startClaudeSession(options = {}) {
-    const timeoutMs = options.timeoutMs ?? 180_000;
-    const effort = options.reasoningEffort ?? "low";
-    const work = await mkdtemp(path.join(options.workRoot ?? tmpdir(), "humanish-claude-session-"));
-    let child;
-    let transport = options.transport;
-    if (transport === undefined) {
-        const spawnFn = options.spawnFn ?? spawn;
-        child = spawnFn("claude", [
-            "-p",
-            "--input-format", "stream-json",
-            "--output-format", "stream-json",
-            // Required for stream-json output in -p mode; it is what makes the per-turn `result` visible.
-            "--verbose",
-            // Read is the only tool it needs — the screenshot — and the only one it gets.
-            "--allowedTools", "Read",
-            ...(options.model === undefined ? [] : ["--model", options.model])
-        ], { cwd: work, stdio: ["pipe", "pipe", "pipe"] });
-        transport = stdioClaudeTransport(child);
-    }
-    let turnIndex = 0;
-    let previousShot;
-    const provider = {
-        id: "local-agent-claude-session",
-        version: options.model ?? "claude (local, operator-authenticated, one session per run)",
-        modelSettings: { reasoningEffort: effort },
-        capabilities: CLAUDE_SESSION_CAPABILITIES,
-        requiresFrame: true,
-        async nextTurn(request, signal) {
-            const frame = request.observation.screenshot;
-            if (frame === undefined) {
-                throw new Error("the Claude session provider needs a screenshot and this observation has none");
-            }
-            turnIndex += 1;
-            const shot = path.join(work, `screen-${String(turnIndex).padStart(3, "0")}.png`);
-            await writeFile(shot, frame);
-            const hint = request.contextHint === undefined ? "" : `\n\nNote from the harness: ${request.contextHint}`;
-            // The persona and the reply shape are in the conversation after turn one; re-sending them
-            // every turn is what the one-shot version had to do, and it is most of what it cost.
-            const text = turnIndex === 1
-                ? promptFor(request, shot, "claude")
-                : `Read the image file ${shot}. That is the CURRENT SCREEN, after your last actions took effect. `
-                    + "Same participant, same task: decide what to do next. "
-                    + "Reply with ONLY a JSON object of the same shape as before."
-                    + hint;
-            transport.send(userMessage(text));
-            const result = await transport.awaitResult(timeoutMs, signal);
-            // The frame it already looked at is not needed on disk; the conversation remembers it.
-            if (previousShot !== undefined)
-                await unlink(previousShot).catch(() => undefined);
-            previousShot = shot;
-            return turnFromResult(result);
-        }
-    };
-    return {
-        provider,
-        close: async () => {
-            transport?.close();
-            child?.kill();
-            await rm(work, { recursive: true, force: true }).catch(() => undefined);
-        }
-    };
-}
-/** Read a stream-json `result` message into a CuaTurn. Exported for tests. */
-export function turnFromResult(result) {
-    if (result.is_error === true || (typeof result.subtype === "string" && result.subtype !== "success")) {
-        // A turn that errored is a BROKEN turn, never an empty one: an empty turn reads to the loop
-        // as "the participant chose to do nothing".
-        throw new Error(`Claude Code turn ended ${String(result.subtype ?? "in error")}: ${String(result.result ?? "").slice(0, 160)}`);
-    }
-    const text = typeof result.result === "string" ? result.result : "";
-    let parsed;
-    try {
-        parsed = parseAgentJson(text);
-    }
-    catch {
-        throw new Error(`Claude Code returned no structured turn output (${text.slice(0, 160)})`);
-    }
-    const actions = toCuaActions(Array.isArray(parsed.actions) ? parsed.actions : []);
-    const done = parsed.done === true || (actions.length === 0 && typeof parsed.message === "string");
-    const outcome = declaredOutcomeOf(parsed.outcome);
-    const usage = result.usage;
-    const count = (key) => (typeof usage?.[key] === "number" ? usage[key] : undefined);
-    const input = count("input_tokens");
-    const output = count("output_tokens");
-    const cachedInput = count("cache_read_input_tokens");
-    const cacheWriteInput = count("cache_creation_input_tokens");
-    return {
-        actions,
-        pendingSafetyChecks: [],
-        done,
-        ...(outcome === undefined ? {} : { outcome }),
-        ...(typeof parsed.reasoning === "string" && parsed.reasoning.length > 0 ? { reasoning: parsed.reasoning } : {}),
-        ...(typeof parsed.message === "string" && parsed.message.length > 0 ? { message: parsed.message } : {}),
-        // Claude Code reports its token counts per turn (#531). They are recorded as counts; the
-        // run's cost line stays "not priced", because a subscription is not a rate card.
-        ...(input === undefined && output === undefined
-            ? {}
-            : {
-                usage: {
-                    ...(input === undefined ? {} : { input }),
-                    ...(output === undefined ? {} : { output }),
-                    ...(cachedInput === undefined ? {} : { cachedInput }),
-                    ...(cacheWriteInput === undefined ? {} : { cacheWriteInput })
-                }
-            })
-    };
-}
-//# sourceMappingURL=local-agent-claude-session.js.map

package/dist/local-agent-claude-session.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"local-agent-claude-session.js","sourceRoot":"","sources":["../src/local-agent-claude-session.ts"],"names":[],"mappings":"AAAA,8FAA8F;AAC9F,EAAE;AACF,gGAAgG;AAChG,8FAA8F;AAC9F,8FAA8F;AAC9F,iGAAiG;AACjG,iGAAiG;AACjG,6FAA6F;AAC7F,EAAE;AACF,0FAA0F;AAC1F,gGAAgG;AAChG,gGAAgG;AAChG,gGAAgG;AAChG,iGAAiG;AACjG,wFAAwF;AACxF,EAAE;AACF,+FAA+F;AAC/F,8BAA8B;AAE9B,OAAO,EAAE,KAAK,EAAuC,MAAM,oBAAoB,CAAC;AAChF,OAAO,EAAE,OAAO,EAAE,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAClE,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AACjC,OAAO,IAAI,MAAM,WAAW,CAAC;AAC7B,OAAO,QAAQ,MAAM,eAAe,CAAC;AAIrC,OAAO,EAAE,iBAAiB,EAAE,cAAc,EAAE,SAAS,EAAE,YAAY,EAAE,wBAAwB,EAAE,MAAM,sBAAsB,CAAC;AAc5H,0GAA0G;AAC1G,MAAM,UAAU,oBAAoB,CAAC,KAAqC;IACxE,MAAM,EAAE,GAAG,QAAQ,CAAC,eAAe,CAAC,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC;IAC7D,IAAI,MAA4F,CAAC;IACjG,IAAI,UAAU,GAAG,EAAE,CAAC;IACpB,IAAI,MAA0B,CAAC;IAE/B,MAAM,IAAI,GAAG,CAAC,KAAY,EAAQ,EAAE;QAClC,MAAM,OAAO,GAAG,MAAM,CAAC;QACvB,MAAM,GAAG,SAAS,CAAC;QACnB,OAAO,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC;IACzB,CAAC,CAAC;IACF,EAAE,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,IAAY,EAAE,EAAE;QAC7B,IAAI,OAAmB,CAAC;QACxB,IAAI,CAAC;YACH,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAe,CAAC;QAC3C,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,CAAC,8BAA8B;QACxC,CAAC;QACD,IAAI,OAAO,CAAC,IAAI,KAAK,QAAQ;YAAE,OAAO;QACtC,MAAM,OAAO,GAAG,MAAM,CAAC;QACvB,MAAM,GAAG,SAAS,CAAC;QACnB,OAAO,EAAE,OAAO,CAAC,OAAO,CAAC,CAAC;IAC5B,CAAC,CAAC,CAAC;IACH,KAAK,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,KAAa,EAAE,EAAE;QACxC,UAAU,GAAG,CAAC,UAAU,GAAG,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC;IACjE,CAAC,CAAC,CAAC;IACH,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,IAAI,EAAE,MAAM,EAAE,EAAE;QACjC,MAAM,GAAG,sBAAsB,IAAI,IAAI,MAAM,MAAM,IAAI,UAAU,EAAE,GAAG,UAAU,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,KAAK,UAAU,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;QAC1H,4FAA4F;QAC5F,iDAAiD;QACjD,IAAI,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC;IAC1B,CAAC,CAAC,CAAC;IAEH,OAAO;QACL,IAAI,CAAC,OAAO;YACV,IAAI,MAAM,KAAK,SAAS;gBAAE,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,CAAC;YAClD,KAAK,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;QACpD,CAAC;QACD,WAAW,CAAC,SAAS,EAAE,MAAM;YAC3B,IAAI,MAAM,KAAK,SAAS;gBAAE,OAAO,OAAO,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC;YACnE,OAAO,IAAI,OAAO,CAAa,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;gBACjD,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,CAAC,yCAAyC,SAAS,IAAI,CAAC,CAAC,EAAE,SAAS,CAAC,CAAC;gBACnH,MAAM,OAAO,GAAG,GAAS,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,CAAC,aAAa,CAAC,CAAC,CAAC;gBAC3D,MAAM,EAAE,gBAAgB,CAAC,OAAO,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;gBAC3D,MAAM,GAAG;oBACP,OAAO,EAAE,CAAC,KAAK,EAAE,EAAE,GAAG,YAAY,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,mBAAmB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;oBAC3G,MAAM,EAAE,CAAC,KAAK,EAAE,EAAE,GAAG,YAAY,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,mBAAmB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;iBAC1G,CAAC;YACJ,CAAC,CAAC,CAAC;QACL,CAAC;QACD,KAAK;YACH,EAAE,CAAC,KAAK,EAAE,CAAC;YACX,KAAK,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC;YAClB,KAAK,CAAC,IAAI,EAAE,CAAC;QACf,CAAC;KACF,CAAC;AACJ,CAAC;AAgBD,MAAM,CAAC,MAAM,2BAA2B,GAAsB,wBAAwB,CAAC;AAQvF,wEAAwE;AACxE,MAAM,UAAU,WAAW,CAAC,IAAY;IACtC,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC,EAAE,EAAE,CAAC;AACxF,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,kBAAkB,CAAC,UAAgC,EAAE;IACzE,MAAM,SAAS,GAAG,OAAO,CAAC,SAAS,IAAI,OAAO,CAAC;IAC/C,MAAM,MAAM,GAAG,OAAO,CAAC,eAAe,IAAI,KAAK,CAAC;IAChD,MAAM,IAAI,GAAG,MAAM,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,QAAQ,IAAI,MAAM,EAAE,EAAE,0BAA0B,CAAC,CAAC,CAAC;IAEhG,IAAI,KAAiD,CAAC;IACtD,IAAI,SAAS,GAAG,OAAO,CAAC,SAAS,CAAC;IAClC,IAAI,SAAS,KAAK,SAAS,EAAE,CAAC;QAC5B,MAAM,OAAO,GAAG,OAAO,CAAC,OAAO,IAAI,KAAK,CAAC;QACzC,KAAK,GAAG,OAAO,CAAC,QAAQ,EAAE;YACxB,IAAI;YACJ,gBAAgB,EAAE,aAAa;YAC/B,iBAAiB,EAAE,aAAa;YAChC,8FAA8F;YAC9F,WAAW;YACX,8EAA8E;YAC9E,gBAAgB,EAAE,MAAM;YACxB,GAAG,CAAC,OAAO,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC;SACnE,EAAE,EAAE,GAAG,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,CAAmC,CAAC;QACrF,SAAS,GAAG,oBAAoB,CAAC,KAAK,CAAC,CAAC;IAC1C,CAAC;IAED,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,YAAgC,CAAC;IACrC,MAAM,QAAQ,GAAgB;QAC5B,EAAE,EAAE,4BAA4B;QAChC,OAAO,EAAE,OAAO,CAAC,KAAK,IAAI,6DAA6D;QACvF,aAAa,EAAE,EAAE,eAAe,EAAE,MAAM,EAAE;QAC1C,YAAY,EAAE,2BAA2B;QACzC,aAAa,EAAE,IAAI;QACnB,KAAK,CAAC,QAAQ,CAAC,OAAuB,EAAE,MAAoB;YAC1D,MAAM,KAAK,GAAG,OAAO,CAAC,WAAW,CAAC,UAAU,CAAC;YAC7C,IAAI,KAAK,KAAK,SAAS,EAAE,CAAC;gBACxB,MAAM,IAAI,KAAK,CAAC,8EAA8E,CAAC,CAAC;YAClG,CAAC;YACD,SAAS,IAAI,CAAC,CAAC;YACf,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,UAAU,MAAM,CAAC,SAAS,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,MAAM,CAAC,CAAC;YACjF,MAAM,SAAS,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;YAC7B,MAAM,IAAI,GAAG,OAAO,CAAC,WAAW,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,8BAA8B,OAAO,CAAC,WAAW,EAAE,CAAC;YAC1G,0FAA0F;YAC1F,qFAAqF;YACrF,MAAM,IAAI,GAAG,SAAS,KAAK,CAAC;gBAC1B,CAAC,CAAC,SAAS,CAAC,OAAO,EAAE,IAAI,EAAE,QAAQ,CAAC;gBACpC,CAAC,CAAC,uBAAuB,IAAI,qEAAqE;sBAC9F,uDAAuD;sBACvD,4DAA4D;sBAC5D,IAAI,CAAC;YACX,SAAU,CAAC,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,CAAC;YACnC,MAAM,MAAM,GAAG,MAAM,SAAU,CAAC,WAAW,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC;YAC/D,uFAAuF;YACvF,IAAI,YAAY,KAAK,SAAS;gBAAE,MAAM,MAAM,CAAC,YAAY,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,SAAS,CAAC,CAAC;YAClF,YAAY,GAAG,IAAI,CAAC;YACpB,OAAO,cAAc,CAAC,MAAM,CAAC,CAAC;QAChC,CAAC;KACF,CAAC;IAEF,OAAO;QACL,QAAQ;QACR,KAAK,EAAE,KAAK,IAAI,EAAE;YAChB,SAAS,EAAE,KAAK,EAAE,CAAC;YACnB,KAAK,EAAE,IAAI,EAAE,CAAC;YACd,MAAM,EAAE,CAAC,IAAI,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,SAAS,CAAC,CAAC;QAC1E,CAAC;KACF,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,MAAM,UAAU,cAAc,CAAC,MAAkB;IAC/C,IAAI,MAAM,CAAC,QAAQ,KAAK,IAAI,IAAI,CAAC,OAAO,MAAM,CAAC,OAAO,KAAK,QAAQ,IAAI,MAAM,CAAC,OAAO,KAAK,SAAS,CAAC,EAAE,CAAC;QACrG,4FAA4F;QAC5F,4CAA4C;QAC5C,MAAM,IAAI,KAAK,CAAC,0BAA0B,MAAM,CAAC,MAAM,CAAC,OAAO,IAAI,UAAU,CAAC,KAAK,MAAM,CAAC,MAAM,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;IAClI,CAAC;IACD,MAAM,IAAI,GAAG,OAAO,MAAM,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC;IACpE,IAAI,MAAkB,CAAC;IACvB,IAAI,CAAC;QACH,MAAM,GAAG,cAAc,CAAC,IAAI,CAAC,CAAC;IAChC,CAAC;IAAC,MAAM,CAAC;QACP,MAAM,IAAI,KAAK,CAAC,mDAAmD,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC;IAC5F,CAAC;IACD,MAAM,OAAO,GAAG,YAAY,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAE,MAAM,CAAC,OAAmB,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;IAC/F,MAAM,IAAI,GAAG,MAAM,CAAC,IAAI,KAAK,IAAI,IAAI,CAAC,OAAO,CAAC,MAAM,KAAK,CAAC,IAAI,OAAO,MAAM,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC;IAClG,MAAM,OAAO,GAAG,iBAAiB,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IAClD,MAAM,KAAK,GAAG,MAAM,CAAC,KAA+B,CAAC;IACrD,MAAM,KAAK,GAAG,CAAC,GAAW,EAAsB,EAAE,CAAC,CAAC,OAAO,KAAK,EAAE,CAAC,GAAG,CAAC,KAAK,QAAQ,CAAC,CAAC,CAAE,KAAK,CAAC,GAAG,CAAY,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC;IAC3H,MAAM,KAAK,GAAG,KAAK,CAAC,cAAc,CAAC,CAAC;IACpC,MAAM,MAAM,GAAG,KAAK,CAAC,eAAe,CAAC,CAAC;IACtC,MAAM,WAAW,GAAG,KAAK,CAAC,yBAAyB,CAAC,CAAC;IACrD,MAAM,eAAe,GAAG,KAAK,CAAC,6BAA6B,CAAC,CAAC;IAC7D,OAAO;QACL,OAAO;QACP,mBAAmB,EAAE,EAAE;QACvB,IAAI;QACJ,GAAG,CAAC,OAAO,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC;QAC7C,GAAG,CAAC,OAAO,MAAM,CAAC,SAAS,KAAK,QAAQ,IAAI,MAAM,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,MAAM,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QAC/G,GAAG,CAAC,OAAO,MAAM,CAAC,OAAO,KAAK,QAAQ,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACvG,yFAAyF;QACzF,iFAAiF;QACjF,GAAG,CAAC,KAAK,KAAK,SAAS,IAAI,MAAM,KAAK,SAAS;YAC7C,CAAC,CAAC,EAAE;YACJ,CAAC,CAAC;gBACE,KAAK,EAAE;oBACL,GAAG,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC;oBACzC,GAAG,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC;oBAC3C,GAAG,CAAC,WAAW,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC;oBACrD,GAAG,CAAC,eAAe,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,CAAC;iBAC9D;aACF,CAAC;KACP,CAAC;AACJ,CAAC"}

package/dist/local-agent-cli.d.ts

@@ -1,121 +0,0 @@
-import type { ActorCapabilities, ParticipantDeclaredOutcome } from "./actor-contract.js";
-import type { CuaAction, CuaProvider, CuaTurnRequest } from "./computer-use.js";
-import type { ReasoningEffort } from "./reasoning-effort.js";
-export type LocalAgentId = "codex" | "claude";
-export interface LocalAgentDescriptor {
-    id: LocalAgentId;
-    /** The command a person types. */
-    bin: string;
-    /** For humans: "Codex (ChatGPT plan)". */
-    label: string;
-    /** Legacy file location; existence is a hint, never an authentication verdict. */
-    credentialPath: string;
-}
-export declare const LOCAL_AGENTS: readonly LocalAgentDescriptor[];
-/**
- * OpenAI structured outputs run in STRICT mode: every property must appear in `required`, so
- * "optional" is expressed as a nullable type. Getting this wrong is a 400 before any thinking
- * happens, which is how this shape was arrived at.
- */
-export declare function localAgentTurnSchema(): Record<string, unknown>;
-interface RawAction {
-    kind?: string;
-    x?: number | null;
-    y?: number | null;
-    text?: string | null;
-    keys?: string[] | null;
-    ms?: number | null;
-}
-/**
- * Map the agent's answer onto the harness action vocabulary. Anything unrecognized is DROPPED
- * rather than guessed at: a coordinate we invented would be recorded as the participant's choice.
- */
-export declare function toCuaActions(raw: readonly RawAction[]): CuaAction[];
-/**
- * Pull the JSON object out of whatever the CLI printed. Codex writes clean JSON to
- * `--output-last-message`; Claude Code wraps it in a ```json fence. Both are handled here rather
- * than in two places, and a response with no object at all is a turn error, never an empty turn —
- * an empty turn would read to the loop as "the participant chose to do nothing".
- */
-/** The declared outcome, only if it is one of the three words; anything else is absence. */
-export declare function declaredOutcomeOf(value: unknown): ParticipantDeclaredOutcome | undefined;
-export declare function parseAgentJson(text: string): Record<string, unknown>;
-export interface SpawnResult {
-    code: number | null;
-    stdout: string;
-    stderr: string;
-}
-export type SpawnLike = (bin: string, args: readonly string[], options: {
-    cwd: string;
-    timeoutMs: number;
-    signal?: AbortSignal;
-}) => Promise<SpawnResult>;
-export interface LocalAgentProviderOptions {
-    agent: LocalAgentId;
-    /** Per-turn wall clock. A coding agent left to think can outlast the run. */
-    timeoutMs?: number;
-    /**
-     * Codex defaults to HIGH effort, which timed out at 240s on a single action; `low` answered the
-     * same screenshot correctly in 9s. A computer-use run is sixty of these, so the default here is
-     * deliberately low and the lab can raise it.
-     */
-    reasoningEffort?: ReasoningEffort;
-    /** Model override passed to the CLI (`--model`). Absent = the CLI's own default. */
-    model?: string;
-    spawnFn?: SpawnLike;
-    /** Scratch root for the screenshot and schema handed to the CLI. */
-    workRoot?: string;
-}
-export declare const LOCAL_AGENT_CAPABILITIES: ActorCapabilities;
-export declare function promptFor(request: CuaTurnRequest, screenshotPath: string, agent: LocalAgentId): string;
-/**
- * A CuaProvider backed by a coding agent that is already signed in on this machine.
- *
- * Deliberately NOT a new lane: the loop, the executor, the trace, the affordance record and the
- * Observer are all unchanged, because the only thing that differs is where the next action comes
- * from. That is also why it is honest to compare a local-agent run against an API run.
- */
-export declare function createLocalAgentProvider(options: LocalAgentProviderOptions): CuaProvider;
-export interface DetectedLocalAgent extends LocalAgentDescriptor {
-    /** Resolved path to the binary. */
-    binPath: string;
-    /**
-     * Whether a credential file exists for it. EXISTENCE ONLY — never read, never parsed, never
-     * reported beyond this boolean. Keyring storage may have no file at all.
-     */
-    credentialsPresent: boolean;
-    /** CLI-reported local status, not a provider request or account-validity test. */
-    authStatus: "authenticated" | "unauthenticated" | "unknown";
-    /** Known Codex login billing class. Absent when the status text is not one of the pinned shapes. */
-    billing?: "account-unknown" | "api";
-}
-export interface DetectLocalAgentsOptions {
-    /** Injected for tests: resolves a binary name to a path, or undefined. */
-    which?: (bin: string) => Promise<string | undefined>;
-    /** Injected for tests: does this path exist? */
-    exists?: (file: string) => Promise<boolean>;
-    home?: string;
-    env?: NodeJS.ProcessEnv;
-    /** Status output is classified in memory and never returned or persisted. */
-    authProbe?: (bin: string, args: readonly string[], env: NodeJS.ProcessEnv) => Promise<SpawnResult>;
-}
-export type HostedCodexCompatibility = "supported" | "unsupported_platform" | "unsupported_version";
-/** Host-only compatibility check. It never initializes app-server or submits a model request. */
-export declare function checkHostedCodexCompatibility(binPath: string, options?: {
-    env?: NodeJS.ProcessEnv;
-    platform?: NodeJS.Platform;
-    arch?: string;
-    probe?: (bin: string, args: readonly string[], env: NodeJS.ProcessEnv) => Promise<SpawnResult>;
-}): Promise<HostedCodexCompatibility>;
-/**
- * Which coding agents are installed and signed in on this machine.
- *
- * This is the whole point of the feature at the surface: someone new does not have to go and make
- * an API key if the thing that can drive the study is already on their laptop. `doctor` says so,
- * and says it as a capability rather than a gate — a machine with no local agent is not broken,
- * it just needs a key.
- */
-export declare function detectLocalAgents(options?: DetectLocalAgentsOptions): Promise<DetectedLocalAgent[]>;
-/** One line for `doctor`, in the register the other rows use. */
-export declare function localAgentDoctorMessage(found: readonly DetectedLocalAgent[]): string;
-export {};