@iowarp/clio-coder 0.4.9 → 0.5.0

package/dist/with-panes-Z6BTEHMZ.js

@@ -1,628 +0,0 @@
-import { createRequire as __clioCreateRequire } from "node:module"; const require = __clioCreateRequire(import.meta.url);
-import {
-  createPanesRuntime
-} from "./chunk-L3NBWLSN.js";
-import {
-  createMuxDomainModule,
-  createYaziSession,
-  resolveYaziBinaries,
-  watchViewerCommand
-} from "./chunk-QB33Z5S7.js";
-import {
-  ensureYaziProfile,
-  yaziProfileDir
-} from "./chunk-JU6MAMRJ.js";
-import "./chunk-QUIJZSDW.js";
-import "./chunk-JABWPKFY.js";
-import "./chunk-XCUDZQLX.js";
-import "./chunk-K2BLQC5E.js";
-import "./chunk-UQIVHSE4.js";
-import "./chunk-LQTKDAUO.js";
-import {
-  sanitizeCallTargetText
-} from "./chunk-XNIZBOFV.js";
-import {
-  BusChannels
-} from "./chunk-W5RM7YAH.js";
-import "./chunk-EFSEIZE6.js";
-import "./chunk-DVNAPPXU.js";
-import "./chunk-KMF54VH3.js";
-import {
-  clioCacheDir,
-  clioConfigDir,
-  clioDataDir,
-  clioStateDir
-} from "./chunk-CTTI7G3T.js";
-import "./chunk-E3LEPLQ2.js";
-import {
-  init_esm_shims
-} from "./chunk-ECBHH7LU.js";
-
-// src/entry/with-panes.ts
-init_esm_shims();
-
-// src/interactive/mux-bridge.ts
-init_esm_shims();
-var MUX_BRIDGE_THROTTLE_MS = 250;
-var LONG_RUN_MS = 6e4;
-var DETAIL_MAX_CHARS = 96;
-function isFailure(outcome) {
-  return outcome === "failed" || outcome === "timed_out";
-}
-function selfStateFor(phase) {
-  if (phase === "tool_blocked") return "blocked";
-  return phase === "idle" || phase === "ended" ? "idle" : "working";
-}
-function terminalOutcome(outcome) {
-  if (outcome === "succeeded") return "succeeded";
-  if (outcome === "canceled") return "canceled";
-  if (outcome === "timed_out") return "timed_out";
-  return "failed";
-}
-function createMuxBridge(deps) {
-  const log = deps.log ?? (() => void 0);
-  const now = deps.now ?? Date.now;
-  const throttleMs = deps.throttleMs ?? MUX_BRIDGE_THROTTLE_MS;
-  const setTimeoutFn = deps.setTimeoutFn ?? setTimeout;
-  const clearTimeoutFn = deps.clearTimeoutFn ?? clearTimeout;
-  const startedAt = /* @__PURE__ */ new Map();
-  let disposed = false;
-  let timer = null;
-  let draining = Promise.resolve();
-  let pendingSelf = null;
-  let reportedSelf = null;
-  const schedule = () => {
-    if (disposed || timer !== null) return;
-    timer = setTimeoutFn(() => {
-      timer = null;
-      void drainSelf();
-    }, throttleMs);
-    timer.unref?.();
-  };
-  const drainSelf = () => {
-    draining = draining.then(async () => {
-      if (disposed) return;
-      const self = pendingSelf;
-      pendingSelf = null;
-      if (self !== null && self !== reportedSelf) {
-        reportedSelf = self;
-        await deps.mux.reportSelf({ state: self }).catch(() => false);
-      }
-    });
-    return draining;
-  };
-  const boundedDetail = (value) => {
-    if (typeof value !== "string" || value.length === 0) return null;
-    return sanitizeCallTargetText(value).slice(0, DETAIL_MAX_CHARS);
-  };
-  const onTerminal = (identity, rawOutcome, fields) => {
-    const outcome = terminalOutcome(rawOutcome);
-    const begun = startedAt.get(identity.runId);
-    startedAt.delete(identity.runId);
-    const policy = deps.notificationsPolicy();
-    const failed = isFailure(outcome);
-    const qualifies = policy === "all" || policy === "failures" && failed;
-    if (!qualifies) return;
-    const detail = boundedDetail(fields.detail);
-    const durationMs = typeof fields.durationMs === "number" ? fields.durationMs : begun === void 0 ? null : now() - begun;
-    if (failed) {
-      deps.notice?.("error", `${identity.agentId} ${outcome} (${identity.runId})${detail ? `: ${detail}` : ""}`);
-    }
-    if (!deps.mux.available()) return;
-    draining = draining.then(async () => {
-      if (disposed) return;
-      try {
-        await deps.mux.notify({
-          title: `${identity.agentId} ${outcome}`,
-          ...detail ? { body: detail } : {},
-          // A failure interrupts; a long success announces. A short success
-          // in `all` mode gets the toast without the chime.
-          sound: failed ? "request" : durationMs !== null && durationMs >= LONG_RUN_MS ? "done" : "none"
-        });
-      } catch (error) {
-        log("debug", `mux bridge notify failed: ${error instanceof Error ? error.message : String(error)}`);
-      }
-    });
-  };
-  const unsubscribers = [
-    deps.bus.on(BusChannels.DispatchStarted, (payload) => {
-      startedAt.set(payload.runId, now());
-    }),
-    deps.bus.on(BusChannels.DispatchCompleted, (payload) => {
-      onTerminal(payload, payload.outcome, { durationMs: payload.durationMs, detail: payload.outcomeDetail });
-    }),
-    deps.bus.on(BusChannels.DispatchFailed, (payload) => {
-      onTerminal(payload, payload.reason, { durationMs: payload.durationMs, detail: payload.outcomeDetail });
-    }),
-    // SA-3. The interactive status machine is the one surface that already
-    // knows when a turn starts, when it ends, and when a tool call is parked
-    // on an approval, so the self-report rides its transitions rather than
-    // growing a second turn-lifecycle listener.
-    deps.bus.on(BusChannels.AgentStatusChanged, (payload) => {
-      const state = selfStateFor(payload.phase);
-      if (state === reportedSelf && pendingSelf === null) return;
-      pendingSelf = state;
-      schedule();
-    })
-  ];
-  return {
-    flush() {
-      if (timer !== null) {
-        clearTimeoutFn(timer);
-        timer = null;
-      }
-      return drainSelf();
-    },
-    dispose() {
-      if (disposed) return;
-      disposed = true;
-      if (timer !== null) clearTimeoutFn(timer);
-      timer = null;
-      for (const unsubscribe of unsubscribers) unsubscribe();
-      startedAt.clear();
-    }
-  };
-}
-
-// src/interactive/watch-pane.ts
-init_esm_shims();
-import { mkdirSync, renameSync, writeFileSync } from "node:fs";
-import { dirname, join } from "node:path";
-function watchSelectionPath(stateDir = clioStateDir()) {
-  return join(stateDir, "watch-selection");
-}
-function atomicWrite(path, content) {
-  mkdirSync(dirname(path), { recursive: true });
-  const tmp = `${path}.tmp`;
-  writeFileSync(tmp, content, "utf8");
-  renameSync(tmp, path);
-}
-function createWatchPaneController(deps) {
-  const selectionPath = deps.selectionPath ?? watchSelectionPath();
-  const dirs = deps.dirs ?? Object.freeze({
-    config: clioConfigDir(),
-    data: clioDataDir(),
-    state: clioStateDir(),
-    cache: clioCacheDir()
-  });
-  const command = deps.command ?? ((path, layout) => watchViewerCommand(path, { dirs: layout }));
-  const write = deps.writeFile ?? atomicWrite;
-  let paneId = null;
-  let disposed = false;
-  let adoptionInFlight = null;
-  const unsubscribe = deps.mux.onPaneGone((record) => {
-    if (record.ref.paneId === paneId) paneId = null;
-  });
-  const adoptExistingPane = () => {
-    if (disposed || paneId !== null) return Promise.resolve(true);
-    if (adoptionInFlight !== null) return adoptionInFlight;
-    const attempt = Promise.resolve().then(() => deps.mux.adoptPane({ purpose: "watch", label: "watch", dock: "workers" })).then((adopted) => {
-      if (!disposed && paneId === null && adopted !== null) paneId = adopted.paneId;
-      return true;
-    }).catch(() => {
-      return false;
-    }).finally(() => {
-      if (adoptionInFlight === attempt) adoptionInFlight = null;
-    });
-    adoptionInFlight = attempt;
-    return attempt;
-  };
-  const writeSelection = (runId) => {
-    try {
-      write(selectionPath, `${runId}
-`);
-      return true;
-    } catch {
-      return false;
-    }
-  };
-  const openPane = (onFailure) => deps.mux.openUtilityPane({
-    ...onFailure ? { onFailure } : {},
-    argv: command(selectionPath, dirs),
-    cwd: deps.getCwd(),
-    label: "watch",
-    title: "clio-coder watch",
-    purpose: "watch",
-    dock: { slot: "workers", ...deps.getWorkersRatio ? { share: deps.getWorkersRatio() } : {} }
-  });
-  void adoptExistingPane();
-  return {
-    async ensureOpen() {
-      if (disposed) return false;
-      if (paneId !== null) return true;
-      await adoptExistingPane();
-      if (paneId !== null) return true;
-      const opened = await openPane();
-      if (opened === null) return false;
-      paneId = opened.paneId;
-      return true;
-    },
-    async watch(runId) {
-      if (!writeSelection(runId)) {
-        return { status: "unavailable", reason: `cannot write the watch selection at ${selectionPath}` };
-      }
-      if (paneId !== null) return { status: "watching", runId, paneId, opened: false };
-      const scanned = await adoptExistingPane();
-      if (!scanned) await adoptExistingPane();
-      if (paneId !== null) return { status: "watching", runId, paneId, opened: false };
-      let reason = "cannot open the watch pane: no failure reason was supplied by the pane host; check /panes before trying again";
-      const opened = await openPane((failure) => {
-        if (failure.source === "local") {
-          reason = `cannot open the watch pane locally: ${failure.message}`;
-        } else {
-          const details = [
-            `kind=${failure.kind}`,
-            ...failure.wireCode === null ? [] : [`code=${failure.wireCode}`],
-            ...failure.method === null ? [] : [`method=${failure.method}`]
-          ];
-          reason = `cannot open the watch pane (${details.join(", ")}): ${failure.message || "no failure message was supplied by the pane host"}`;
-        }
-      });
-      if (opened === null) {
-        return { status: "unavailable", reason };
-      }
-      paneId = opened.paneId;
-      return { status: "watching", runId, paneId, opened: true };
-    },
-    follow(runId) {
-      if (paneId === null) return false;
-      return writeSelection(runId);
-    },
-    isOpen() {
-      return paneId !== null;
-    },
-    dispose() {
-      disposed = true;
-      unsubscribe();
-    }
-  };
-}
-
-// src/interactive/yazi-bridge.ts
-init_esm_shims();
-import { spawnSync } from "node:child_process";
-import { randomUUID } from "node:crypto";
-import { mkdirSync as mkdirSync2, readFileSync, statSync, writeFileSync as writeFileSync2 } from "node:fs";
-import { isAbsolute, join as join2, relative, resolve } from "node:path";
-var YAZI_PICK_MAX_PATHS = 32;
-var YAZI_PICK_MAX_CHARS = 4096;
-var YAZI_LIVENESS_MS = 5e3;
-function readOptional(path) {
-  try {
-    return readFileSync(path, "utf8");
-  } catch {
-    return "";
-  }
-}
-async function runYaziTerminalChooser(options) {
-  const binaries = (options.resolveBinaries ?? resolveYaziBinaries)();
-  if (!binaries.yaziPath) {
-    return { status: "missing-binary", binary: "yazi", detail: binaries.missingYaziDetail };
-  }
-  if (options.profileMode === "managed" && !binaries.yaPath) {
-    return {
-      status: "missing-binary",
-      binary: "ya",
-      detail: "the Yazi helper `ya` is not available; install the complete pinned Yazi tool"
-    };
-  }
-  const cacheDir = options.cacheDir ?? clioCacheDir();
-  const sessionsDir = join2(cacheDir, "yazi", "sessions");
-  mkdirSync2(sessionsDir, { recursive: true });
-  const sessionId = options.sessionId ?? randomUUID();
-  const chooserPath = join2(sessionsDir, `${sessionId}.chooser`);
-  const cwdPath = join2(sessionsDir, `${sessionId}.cwd`);
-  const profile = options.profileMode === "managed" ? (options.ensureProfile ?? ensureYaziProfile)({
-    yaPath: binaries.yaPath,
-    yaziPath: binaries.yaziPath,
-    profileDir: yaziProfileDir(cacheDir)
-  }) : null;
-  if (options.profileMode === "managed" && !profile) {
-    return { status: "profile-error", reason: "Clio could not generate a valid managed Yazi profile" };
-  }
-  writeFileSync2(chooserPath, "");
-  writeFileSync2(cwdPath, "");
-  const args = [options.cwd, "--chooser-file", chooserPath, "--cwd-file", cwdPath];
-  const env = { ...process.env, ...profile ? { YAZI_CONFIG_HOME: profile.dir } : {} };
-  let processError;
-  options.stopUi();
-  try {
-    const result = options.spawnYazi ? options.spawnYazi(binaries.yaziPath, args, { cwd: options.cwd, env }) : spawnSync(binaries.yaziPath, args, { cwd: options.cwd, env, stdio: "inherit" });
-    processError = result.error;
-  } catch (error) {
-    processError = error instanceof Error ? error : new Error(String(error));
-  } finally {
-    options.startUi();
-    options.requestRender();
-  }
-  if (processError) return { status: "unavailable", reason: `yazi chooser failed: ${processError.message}` };
-  const paths = readOptional(chooserPath).split("\n").filter((path) => path.length > 0);
-  if (paths.length === 0) return { status: "cancelled" };
-  return {
-    status: "chosen",
-    choice: { paths, cwd: readOptional(cwdPath).trim() || options.cwd }
-  };
-}
-function containsAsciiWhitespace(value) {
-  for (const character of value) {
-    const code = character.charCodeAt(0);
-    if (code === 32 || code >= 9 && code <= 13) return true;
-  }
-  return false;
-}
-function relativeWhenInside(path, cwd) {
-  const candidate = relative(cwd, path);
-  if (candidate.length > 0 && candidate !== ".." && !candidate.startsWith(`..${process.platform === "win32" ? "\\" : "/"}`) && !isAbsolute(candidate)) {
-    return candidate;
-  }
-  return path;
-}
-function draftContains(draft, item) {
-  let from = 0;
-  while (from <= draft.length) {
-    const at = draft.indexOf(item, from);
-    if (at < 0) return false;
-    const before = at === 0 ? "" : draft[at - 1] ?? "";
-    const after = draft[at + item.length] ?? "";
-    if ((before === "" || /\s/u.test(before)) && (after === "" || /\s/u.test(after))) return true;
-    from = at + item.length;
-  }
-  return false;
-}
-function pathKind(path) {
-  try {
-    const stat = statSync(path);
-    if (stat.isFile()) return "file";
-    if (stat.isDirectory()) return "directory";
-    return "other";
-  } catch {
-    return "other";
-  }
-}
-function formatYaziMentions(paths, options) {
-  const maxPaths = options.maxPaths ?? YAZI_PICK_MAX_PATHS;
-  const maxChars = options.maxChars ?? YAZI_PICK_MAX_CHARS;
-  const inspect = options.statPath ?? pathKind;
-  const rendered = [];
-  const seen = /* @__PURE__ */ new Set();
-  let mentions = 0;
-  let plainPaths = 0;
-  let duplicates = 0;
-  let truncated = Math.max(0, paths.length - maxPaths);
-  const candidates = paths.slice(0, maxPaths);
-  for (const [index, rawPath] of candidates.entries()) {
-    const absolute = isAbsolute(rawPath) ? rawPath : resolve(options.cwd, rawPath);
-    if (seen.has(absolute)) {
-      duplicates += 1;
-      continue;
-    }
-    seen.add(absolute);
-    const kind = inspect(absolute);
-    let display = relativeWhenInside(absolute, options.cwd);
-    if (kind === "directory" && display !== "/" && !display.endsWith("/")) display = `${display}/`;
-    const mention = kind === "file" && !containsAsciiWhitespace(display);
-    const item = mention ? `@${display}` : `\`${display.replaceAll("`", "\\`")}\``;
-    if (draftContains(options.draft, item)) {
-      duplicates += 1;
-      continue;
-    }
-    const nextLength = rendered.length === 0 ? item.length : rendered.join(" ").length + 1 + item.length;
-    if (nextLength > maxChars) {
-      truncated += candidates.length - index;
-      break;
-    }
-    rendered.push(item);
-    if (mention) mentions += 1;
-    else plainPaths += 1;
-  }
-  return {
-    text: rendered.join(" "),
-    inserted: rendered.length,
-    mentions,
-    plainPaths,
-    duplicates,
-    truncated
-  };
-}
-function appendDraft(draft, addition) {
-  if (addition.length === 0) return draft;
-  if (draft.length === 0) return addition;
-  return `${draft.replace(/\s+$/u, "")} ${addition}`;
-}
-function createYaziBridge(deps) {
-  const openSession = deps.openSession ?? createYaziSession;
-  let active = null;
-  let paneCwd = null;
-  let droppedLines = 0;
-  let disposed = false;
-  let generation = 0;
-  let livenessTimer = null;
-  let sawLine = false;
-  const clearLiveness = () => {
-    if (livenessTimer) clearTimeout(livenessTimer);
-    livenessTimer = null;
-  };
-  const insert = (paths, cwd) => {
-    const draft = deps.getDraft();
-    const batch = formatYaziMentions(paths, { cwd, draft, ...deps.statPath ? { statPath: deps.statPath } : {} });
-    if (batch.inserted === 0) {
-      if (batch.truncated > 0) {
-        deps.notice(
-          "warning",
-          `${batch.truncated} file-pane path${batch.truncated === 1 ? " was" : "s were"} omitted by the pick limits`
-        );
-      }
-      return;
-    }
-    deps.setDraft(appendDraft(draft, batch.text));
-    deps.requestRender();
-    void deps.mux?.focusSelf();
-    deps.notice(
-      "info",
-      `${batch.inserted} ${batch.inserted === 1 ? "path" : "paths"} from the files pane added to the draft`
-    );
-    if (batch.plainPaths > 0) {
-      deps.notice(
-        "info",
-        `${batch.plainPaths} ${batch.plainPaths === 1 ? "path was" : "paths were"} inserted as plain text because directories and spaced paths cannot be file mentions`
-      );
-    }
-    if (batch.truncated > 0) {
-      deps.notice(
-        "warning",
-        `${batch.truncated} file-pane path${batch.truncated === 1 ? " was" : "s were"} omitted by the pick limits`
-      );
-    }
-  };
-  const start = async (mode, profileMode, dockShare) => {
-    const mux = deps.mux;
-    if (!mux?.available()) return { status: "unavailable", reason: "the pane layer is not available" };
-    generation += 1;
-    const ownGeneration = generation;
-    const token = (deps.pickToken ?? randomUUID)();
-    sawLine = false;
-    paneCwd = deps.getCwd();
-    const result = await openSession({
-      mux,
-      mode,
-      profileMode,
-      cwd: deps.getCwd(),
-      ...dockShare === void 0 ? {} : { dockShare },
-      pickToken: token,
-      onEvent: (event) => {
-        if (disposed || ownGeneration !== generation) return;
-        sawLine = true;
-        if (event.kind === "cd") {
-          paneCwd = event.cwd;
-          return;
-        }
-        if (event.values[0] !== token) {
-          droppedLines += 1;
-          return;
-        }
-        insert(event.values.slice(1), paneCwd ?? deps.getCwd());
-      },
-      onChooser: (choice) => {
-        if (disposed || ownGeneration !== generation) return;
-        paneCwd = choice.cwd;
-        if (choice.paths.length > 0) insert(choice.paths, choice.cwd);
-        else void deps.mux?.focusSelf();
-      },
-      onStopped: () => {
-        if (ownGeneration !== generation) return;
-        clearLiveness();
-        active = null;
-      }
-    });
-    if (result.status !== "opened") return result;
-    if (disposed || ownGeneration !== generation) {
-      await result.session.close();
-      return { status: "unavailable", reason: "the file pane was closed while it was opening" };
-    }
-    active = result.session;
-    await mux.focusPane(result.session.pane.paneId);
-    if (mode === "companion") {
-      livenessTimer = setTimeout(() => {
-        if (disposed || ownGeneration !== generation || sawLine || active === null) return;
-        const stale = active;
-        active = null;
-        deps.notice("warning", "the files pane did not report back in time; reopening it in pick mode");
-        void stale.close().finally(() => {
-          if (!disposed && ownGeneration === generation) void start("chooser", profileMode, dockShare);
-        });
-      }, deps.livenessMs ?? YAZI_LIVENESS_MS);
-      livenessTimer.unref();
-    }
-    return { status: "opened", mode, paneId: result.session.pane.paneId, existing: false };
-  };
-  const liveSession = () => {
-    if (active === null) return null;
-    const mux = deps.mux;
-    if (mux?.available() && !mux.list().some((record) => record.ref.paneId === active?.pane.paneId)) {
-      active = null;
-      return null;
-    }
-    return active;
-  };
-  const closeActive = async () => {
-    const session = liveSession();
-    if (session === null) return false;
-    generation += 1;
-    clearLiveness();
-    active = null;
-    await session.close();
-    await deps.mux?.focusSelf();
-    return true;
-  };
-  return {
-    async open(options = {}) {
-      if (disposed) return { status: "unavailable", reason: "the files pane bridge is closed" };
-      const settings = deps.getSettings?.() ?? { mode: "companion", profile: "managed", followCwd: true };
-      const existing = liveSession();
-      if (existing) {
-        if (settings.followCwd) {
-          const cwd = deps.getCwd();
-          if (await existing.pushCwd(cwd)) paneCwd = cwd;
-        }
-        await deps.mux?.unzoomSelf();
-        await deps.mux?.focusPane(existing.pane.paneId);
-        return { status: "opened", mode: existing.mode, paneId: existing.pane.paneId, existing: true };
-      }
-      const mode = options.once || settings.profile === "user" ? "chooser" : settings.mode;
-      if (!deps.mux?.available()) {
-        const result = await (deps.runTerminalChooser ?? ((choiceOptions) => runYaziTerminalChooser({
-          ...choiceOptions,
-          stopUi: deps.stopUi ?? (() => {
-          }),
-          startUi: deps.startUi ?? (() => {
-          }),
-          requestRender: deps.requestRender
-        })))({ cwd: deps.getCwd(), profileMode: settings.profile });
-        if (result.status === "chosen") {
-          paneCwd = result.choice.cwd;
-          insert(result.choice.paths, result.choice.cwd);
-        }
-        if (result.status === "chosen" || result.status === "cancelled") {
-          return { status: "opened", mode: "chooser", paneId: null, existing: false };
-        }
-        return result;
-      }
-      await deps.mux.unzoomSelf();
-      return await start(mode, settings.profile, settings.ratio);
-    },
-    close() {
-      if (disposed) return Promise.resolve(false);
-      return closeActive();
-    },
-    isOpen() {
-      return liveSession() !== null;
-    },
-    status() {
-      const snapshot = active?.snapshot();
-      return {
-        mode: active?.mode ?? "closed",
-        paneId: active?.pane.paneId ?? null,
-        paneCwd: paneCwd ?? snapshot?.paneCwd ?? null,
-        lastLineAt: snapshot?.lastLineAt ?? null,
-        droppedLines
-      };
-    },
-    dispose() {
-      if (disposed) return;
-      disposed = true;
-      generation += 1;
-      clearLiveness();
-      const closing = active;
-      active = null;
-      if (closing) void closing.close();
-    }
-  };
-}
-export {
-  createMuxBridge,
-  createMuxDomainModule,
-  createPanesRuntime,
-  createWatchPaneController,
-  createYaziBridge
-};
-//# sourceMappingURL=with-panes-Z6BTEHMZ.js.map

package/evals/README.md

@@ -1,9 +0,0 @@
-# Reference evals
-
-These suites exercise Clio's one evaluation engine from a source checkout: run the deterministic built-in authority corpus with `node dist/cli/index.js eval run --suite evals/behavioral-machinery.yaml --clio-coder-entry dist/cli/index.js`, or select a configured model target for `behavioral-model.yaml`, `behavioral-model-negative-control.yaml`, and `tracked-metrics-baseline.yaml`. The npm archive keeps these inputs for inspection and reproducibility, but the deterministic TypeScript driver uses the repository development toolchain. The graders and fixtures beside the suites are versioned inputs; `behavioral-machinery-baseline.json` is retained as reference evidence, not as a release gate.
-
-The single-system boundary is the engine under `src/domains/eval`: suites, verdict validation, artifact envelopes, comparison, and reporting belong there, while `evals/` contains source-checkout reference inputs and deterministic drivers. In v0.5.0, external-benchmark adapters such as SWE-bench and Terminal-Bench should translate their cases and grader observations into this engine's suite, execution, and verdict contracts, then consume its artifacts and comparison reports instead of adding a parallel harness.
-
-`tool-surface-model.yaml` is a dedicated model-required suite for read tail arguments, bash working-directory reset between calls, and rejection of a non-HTTP web_fetch URL. Run it with a configured target using `node dist/cli/index.js eval run --suite evals/tool-surface-model.yaml --target <target-id> --model <model-id> --trials 3 --clio-coder-entry dist/cli/index.js`. This invokes the selected model and incurs its normal inference cost. The URL case deliberately expects one tool error; it requires no successful web request. The target must expose all three tools, and local policy blocks are measured separately from execution errors.
-
-Each trial records `tools.calls.<tool>`, `tools.succeeded.<tool>`, `tools.failed.<tool>`, and `tools.blocked.<tool>` from terminal tool events, plus grader correctness. Compare repeated runs with the same target, model, and settings when evaluating a description change; no arbitrary text perturbation is guaranteed to worsen model behavior. CI executes the suite and real graders against controlled successful and failing event fixtures without calling a model. Those contracts validate the evaluation machinery and are not live model scores.