opensquid 0.5.395

package/dist/hooks/honesty-ledger.js

@@ -0,0 +1,226 @@
+/**
+ * Honesty ledger — catches claim-vs-action gaps in assistant turns.
+ *
+ * The agent makes claims like "running tests now" or "starting research"
+ * or "committed" in its text output. These claims are checkable against
+ * the tool calls that actually happened in the same turn. When the
+ * claim has no matching tool call, that's a "broken promise" — opensquid
+ * records it so the next turn surfaces it to the agent for correction.
+ *
+ * Storage:
+ *   <data-root>/sessions/<session-id>/turn-ledger.jsonl
+ *     One JSON line per tool call this turn, appended by the PreToolUse
+ *     hook. Cleared at turn-end after reconciliation.
+ *
+ *   <data-root>/sessions/<session-id>/broken-promises.jsonl
+ *     Append-only ledger of claims that lacked matching evidence.
+ *     Surfaced to the agent on the NEXT turn via SessionStart/
+ *     UserPromptSubmit hook output.
+ */
+import { promises as fs } from "node:fs";
+import * as path from "node:path";
+import { loadBundledDefaultCodex } from "../codex/loader.js";
+import { resolveDataRoot } from "../codex/store.js";
+/**
+ * Bridge: convert a codex CodexClaimEvidence into the ledger-side
+ * ClaimEvidenceShape. The two shapes overlap but differ in two ways:
+ * codex uses `tool_call` while the ledger uses `tool_called` (a
+ * legacy past-participle name); and codex `input_contains` carries
+ * an extra `field` parameter that the ledger doesn't use.
+ *
+ * The bridge is the single seam where the two vocabularies are
+ * reconciled — keeping it in one place means the rest of the file
+ * can keep using its existing `ClaimEvidenceShape`.
+ */
+function codexEvidenceToLedgerEvidence(c) {
+    switch (c.kind) {
+        case "any_tool":
+            return { kind: "any_tool" };
+        case "tool_call":
+            return { kind: "tool_called", tool: c.tool };
+        case "bash_contains":
+            return { kind: "bash_contains", needle: c.needle };
+        case "bash_regex":
+            return { kind: "bash_regex", pattern: c.pattern };
+        case "input_contains":
+            // Drop `field` — the ledger's input_contains matches against
+            // the whole input_summary, not a specific field.
+            return { kind: "input_contains", tool: c.tool, needle: c.needle };
+        case "any_of":
+            return {
+                kind: "any_of",
+                options: c.options.map(codexEvidenceToLedgerEvidence),
+            };
+    }
+}
+function codexClaimToLedgerPattern(c) {
+    return {
+        id: c.id,
+        text_regex: c.claim_pattern,
+        evidence: codexEvidenceToLedgerEvidence(c.evidence),
+        promise_label: c.unfulfilled_message,
+    };
+}
+/**
+ * Claim catalog — sourced from the bundled-default codex (0.7.17,
+ * drift-as-codex chunk 3b). Previously a hand-maintained TS array;
+ * now loaded once at module init from
+ * `src/codex/bundled-default/codex.yaml` via the chunk-2 loader.
+ *
+ * Fail-open: if the codex is unloadable, the catalog is empty and
+ * no claims fire. Better silent under-enforcement than a hook crash.
+ */
+export const CLAIM_PATTERNS = (() => {
+    try {
+        const codex = loadBundledDefaultCodex();
+        return (codex.claims ?? []).map(codexClaimToLedgerPattern);
+    }
+    catch (err) {
+        process.stderr.write(`[opensquid honesty-ledger] codex unloadable, claim catalog empty: ${err instanceof Error ? err.message : err}\n`);
+        return [];
+    }
+})();
+function sessionDir(sessionId, dataRoot) {
+    return path.join(resolveDataRoot(dataRoot), "sessions", sessionId);
+}
+function ledgerPath(sessionId, dataRoot) {
+    // Filename kept as `turn-ledger.jsonl` for back-compat with any
+    // existing per-turn files on disk; semantics are now session-scoped.
+    return path.join(sessionDir(sessionId, dataRoot), "turn-ledger.jsonl");
+}
+function brokenPromisesPath(sessionId, dataRoot) {
+    return path.join(sessionDir(sessionId, dataRoot), "broken-promises.jsonl");
+}
+/** Called by PreToolUse hook to record what the agent is about to do. */
+export async function recordToolCall(sessionId, tool, inputSummary, options = {}) {
+    const dir = sessionDir(sessionId, options.dataRoot);
+    await fs.mkdir(dir, { recursive: true });
+    const entry = {
+        ts: new Date().toISOString(),
+        tool,
+        input_summary: inputSummary,
+    };
+    await fs.appendFile(ledgerPath(sessionId, options.dataRoot), JSON.stringify(entry) + "\n", "utf8");
+}
+/** Read the session-scoped ledger (every tool call so far this session). */
+export async function readTurnLedger(sessionId, options = {}) {
+    try {
+        const raw = await fs.readFile(ledgerPath(sessionId, options.dataRoot), "utf8");
+        return raw
+            .split("\n")
+            .filter((line) => line.trim())
+            .map((line) => JSON.parse(line));
+    }
+    catch {
+        return [];
+    }
+}
+/**
+ * Clear the session ledger. ONLY called at session end (or by tests).
+ * Stop hook does NOT call this anymore — the ledger persists across
+ * turns to avoid recap-text false-positives.
+ */
+export async function clearTurnLedger(sessionId, options = {}) {
+    try {
+        await fs.rm(ledgerPath(sessionId, options.dataRoot));
+    }
+    catch {
+        // already gone, fine
+    }
+}
+/**
+ * Explicit session-end clear: wipes everything opensquid wrote under
+ * this session's directory — turn ledger, broken promises, plus the
+ * heartbeat checkpoint and pending marker (#124). Files are removed
+ * individually so unrelated files in the session dir survive (in case
+ * a future hook drops something else there).
+ */
+export async function clearSession(sessionId, options = {}) {
+    const sessionFiles = [
+        ledgerPath(sessionId, options.dataRoot),
+        brokenPromisesPath(sessionId, options.dataRoot),
+        path.join(sessionDir(sessionId, options.dataRoot), "heartbeat-checkpoint.json"),
+        path.join(sessionDir(sessionId, options.dataRoot), "heartbeat-pending.txt"),
+    ];
+    for (const p of sessionFiles) {
+        try {
+            await fs.rm(p);
+        }
+        catch {
+            // ignore
+        }
+    }
+}
+/**
+ * Scan assistant text for claim phrases, reconcile against the ledger,
+ * return any unfulfilled promises.
+ */
+export function reconcile(assistantText, ledger) {
+    const broken = [];
+    for (const claim of CLAIM_PATTERNS) {
+        let re;
+        try {
+            re = new RegExp(claim.text_regex, "i");
+        }
+        catch {
+            continue;
+        }
+        const m = assistantText.match(re);
+        if (!m)
+            continue;
+        if (hasEvidence(claim.evidence, ledger))
+            continue;
+        broken.push({
+            ts: new Date().toISOString(),
+            claim_id: claim.id,
+            claim_label: claim.promise_label,
+            matched_text: m[0],
+            reason: `assistant claimed "${m[0]}" but no matching tool call in this turn`,
+        });
+    }
+    return broken;
+}
+function hasEvidence(ev, ledger) {
+    switch (ev.kind) {
+        case "any_tool":
+            return ledger.length > 0;
+        case "tool_called":
+            return ledger.some((e) => e.tool === ev.tool);
+        case "bash_contains":
+            return ledger.some((e) => e.tool === "Bash" && e.input_summary.includes(ev.needle));
+        case "bash_regex": {
+            let re;
+            try {
+                re = new RegExp(ev.pattern);
+            }
+            catch {
+                return true; // bad pattern → fail open, don't false-positive a broken promise
+            }
+            return ledger.some((e) => e.tool === "Bash" && re.test(e.input_summary));
+        }
+        case "any_of":
+            return ev.options.some((opt) => hasEvidence(opt, ledger));
+        case "input_contains":
+            return ledger.some((e) => e.tool === ev.tool && e.input_summary.includes(ev.needle));
+    }
+}
+/** Append a broken promise to the session's append-only ledger. */
+export async function recordBrokenPromise(sessionId, promise, options = {}) {
+    const dir = sessionDir(sessionId, options.dataRoot);
+    await fs.mkdir(dir, { recursive: true });
+    await fs.appendFile(brokenPromisesPath(sessionId, options.dataRoot), JSON.stringify(promise) + "\n", "utf8");
+}
+/** Read all broken promises for a session (used by next turn's hook). */
+export async function readBrokenPromises(sessionId, options = {}) {
+    try {
+        const raw = await fs.readFile(brokenPromisesPath(sessionId, options.dataRoot), "utf8");
+        return raw
+            .split("\n")
+            .filter((line) => line.trim())
+            .map((line) => JSON.parse(line));
+    }
+    catch {
+        return [];
+    }
+}
+//# sourceMappingURL=honesty-ledger.js.map

package/dist/hooks/honesty-ledger.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"honesty-ledger.js","sourceRoot":"","sources":["../../src.legacy/hooks/honesty-ledger.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAEH,OAAO,EAAE,QAAQ,IAAI,EAAE,EAAE,MAAM,SAAS,CAAC;AACzC,OAAO,KAAK,IAAI,MAAM,WAAW,CAAC;AAElC,OAAO,EAAE,uBAAuB,EAAE,MAAM,oBAAoB,CAAC;AAC7D,OAAO,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AA8BpD;;;;;;;;;;GAUG;AACH,SAAS,6BAA6B,CAAC,CAAqB;IAC1D,QAAQ,CAAC,CAAC,IAAI,EAAE,CAAC;QACf,KAAK,UAAU;YACb,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC;QAC9B,KAAK,WAAW;YACd,OAAO,EAAE,IAAI,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QAC/C,KAAK,eAAe;YAClB,OAAO,EAAE,IAAI,EAAE,eAAe,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC;QACrD,KAAK,YAAY;YACf,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC;QACpD,KAAK,gBAAgB;YACnB,6DAA6D;YAC7D,iDAAiD;YACjD,OAAO,EAAE,IAAI,EAAE,gBAAgB,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC;QACpE,KAAK,QAAQ;YACX,OAAO;gBACL,IAAI,EAAE,QAAQ;gBACd,OAAO,EAAE,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,6BAA6B,CAAC;aACtD,CAAC;IACN,CAAC;AACH,CAAC;AAED,SAAS,yBAAyB,CAAC,CAAkB;IACnD,OAAO;QACL,EAAE,EAAE,CAAC,CAAC,EAAE;QACR,UAAU,EAAE,CAAC,CAAC,aAAa;QAC3B,QAAQ,EAAE,6BAA6B,CAAC,CAAC,CAAC,QAAQ,CAAC;QACnD,aAAa,EAAE,CAAC,CAAC,mBAAmB;KACrC,CAAC;AACJ,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,CAAC,MAAM,cAAc,GAAmB,CAAC,GAAG,EAAE;IAClD,IAAI,CAAC;QACH,MAAM,KAAK,GAAG,uBAAuB,EAAE,CAAC;QACxC,OAAO,CAAC,KAAK,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,yBAAyB,CAAC,CAAC;IAC7D,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,OAAO,CAAC,MAAM,CAAC,KAAK,CAClB,qEAAqE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,IAAI,CAClH,CAAC;QACF,OAAO,EAAE,CAAC;IACZ,CAAC;AACH,CAAC,CAAC,EAAE,CAAC;AAyBL,SAAS,UAAU,CAAC,SAAiB,EAAE,QAAiB;IACtD,OAAO,IAAI,CAAC,IAAI,CAAC,eAAe,CAAC,QAAQ,CAAC,EAAE,UAAU,EAAE,SAAS,CAAC,CAAC;AACrE,CAAC;AAED,SAAS,UAAU,CAAC,SAAiB,EAAE,QAAiB;IACtD,gEAAgE;IAChE,qEAAqE;IACrE,OAAO,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,SAAS,EAAE,QAAQ,CAAC,EAAE,mBAAmB,CAAC,CAAC;AACzE,CAAC;AAED,SAAS,kBAAkB,CAAC,SAAiB,EAAE,QAAiB;IAC9D,OAAO,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,SAAS,EAAE,QAAQ,CAAC,EAAE,uBAAuB,CAAC,CAAC;AAC7E,CAAC;AAED,yEAAyE;AACzE,MAAM,CAAC,KAAK,UAAU,cAAc,CAClC,SAAiB,EACjB,IAAY,EACZ,YAAoB,EACpB,UAAiC,EAAE;IAEnC,MAAM,GAAG,GAAG,UAAU,CAAC,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC,CAAC;IACpD,MAAM,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACzC,MAAM,KAAK,GAAoB;QAC7B,EAAE,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;QAC5B,IAAI;QACJ,aAAa,EAAE,YAAY;KAC5B,CAAC;IACF,MAAM,EAAE,CAAC,UAAU,CACjB,UAAU,CAAC,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC,EACvC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,IAAI,EAC5B,MAAM,CACP,CAAC;AACJ,CAAC;AAED,4EAA4E;AAC5E,MAAM,CAAC,KAAK,UAAU,cAAc,CAClC,SAAiB,EACjB,UAAiC,EAAE;IAEnC,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,MAAM,EAAE,CAAC,QAAQ,CAAC,UAAU,CAAC,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC,CAAC;QAC/E,OAAO,GAAG;aACP,KAAK,CAAC,IAAI,CAAC;aACX,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC;aAC7B,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAoB,CAAC,CAAC;IACxD,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,CAAC;IACZ,CAAC;AACH,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,eAAe,CACnC,SAAiB,EACjB,UAAiC,EAAE;IAEnC,IAAI,CAAC;QACH,MAAM,EAAE,CAAC,EAAE,CAAC,UAAU,CAAC,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC,CAAC,CAAC;IACvD,CAAC;IAAC,MAAM,CAAC;QACP,qBAAqB;IACvB,CAAC;AACH,CAAC;AAED;;;;;;GAMG;AACH,MAAM,CAAC,KAAK,UAAU,YAAY,CAChC,SAAiB,EACjB,UAAiC,EAAE;IAEnC,MAAM,YAAY,GAAG;QACnB,UAAU,CAAC,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC;QACvC,kBAAkB,CAAC,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC;QAC/C,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC,EAAE,2BAA2B,CAAC;QAC/E,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC,EAAE,uBAAuB,CAAC;KAC5E,CAAC;IACF,KAAK,MAAM,CAAC,IAAI,YAAY,EAAE,CAAC;QAC7B,IAAI,CAAC;YACH,MAAM,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;QACjB,CAAC;QAAC,MAAM,CAAC;YACP,SAAS;QACX,CAAC;IACH,CAAC;AACH,CAAC;AAcD;;;GAGG;AACH,MAAM,UAAU,SAAS,CAAC,aAAqB,EAAE,MAAyB;IACxE,MAAM,MAAM,GAAoB,EAAE,CAAC;IACnC,KAAK,MAAM,KAAK,IAAI,cAAc,EAAE,CAAC;QACnC,IAAI,EAAU,CAAC;QACf,IAAI,CAAC;YACH,EAAE,GAAG,IAAI,MAAM,CAAC,KAAK,CAAC,UAAU,EAAE,GAAG,CAAC,CAAC;QACzC,CAAC;QAAC,MAAM,CAAC;YACP,SAAS;QACX,CAAC;QACD,MAAM,CAAC,GAAG,aAAa,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QAClC,IAAI,CAAC,CAAC;YAAE,SAAS;QACjB,IAAI,WAAW,CAAC,KAAK,CAAC,QAAQ,EAAE,MAAM,CAAC;YAAE,SAAS;QAClD,MAAM,CAAC,IAAI,CAAC;YACV,EAAE,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YAC5B,QAAQ,EAAE,KAAK,CAAC,EAAE;YAClB,WAAW,EAAE,KAAK,CAAC,aAAa;YAChC,YAAY,EAAE,CAAC,CAAC,CAAC,CAAC;YAClB,MAAM,EAAE,sBAAsB,CAAC,CAAC,CAAC,CAAC,0CAA0C;SAC7E,CAAC,CAAC;IACL,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,SAAS,WAAW,CAAC,EAAsB,EAAE,MAAyB;IACpE,QAAQ,EAAE,CAAC,IAAI,EAAE,CAAC;QAChB,KAAK,UAAU;YACb,OAAO,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC;QAC3B,KAAK,aAAa;YAChB,OAAO,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,EAAE,CAAC,IAAI,CAAC,CAAC;QAChD,KAAK,eAAe;YAClB,OAAO,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,IAAI,CAAC,CAAC,aAAa,CAAC,QAAQ,CAAC,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC;QACtF,KAAK,YAAY,CAAC,CAAC,CAAC;YAClB,IAAI,EAAU,CAAC;YACf,IAAI,CAAC;gBACH,EAAE,GAAG,IAAI,MAAM,CAAC,EAAE,CAAC,OAAO,CAAC,CAAC;YAC9B,CAAC;YAAC,MAAM,CAAC;gBACP,OAAO,IAAI,CAAC,CAAC,iEAAiE;YAChF,CAAC;YACD,OAAO,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,IAAI,EAAE,CAAC,IAAI,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC;QAC3E,CAAC;QACD,KAAK,QAAQ;YACX,OAAO,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC,CAAC;QAC5D,KAAK,gBAAgB;YACnB,OAAO,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,EAAE,CAAC,IAAI,IAAI,CAAC,CAAC,aAAa,CAAC,QAAQ,CAAC,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC;IACzF,CAAC;AACH,CAAC;AAED,mEAAmE;AACnE,MAAM,CAAC,KAAK,UAAU,mBAAmB,CACvC,SAAiB,EACjB,OAAsB,EACtB,UAAiC,EAAE;IAEnC,MAAM,GAAG,GAAG,UAAU,CAAC,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC,CAAC;IACpD,MAAM,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACzC,MAAM,EAAE,CAAC,UAAU,CACjB,kBAAkB,CAAC,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC,EAC/C,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,GAAG,IAAI,EAC9B,MAAM,CACP,CAAC;AACJ,CAAC;AAED,yEAAyE;AACzE,MAAM,CAAC,KAAK,UAAU,kBAAkB,CACtC,SAAiB,EACjB,UAAiC,EAAE;IAEnC,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,MAAM,EAAE,CAAC,QAAQ,CAAC,kBAAkB,CAAC,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC,CAAC;QACvF,OAAO,GAAG;aACP,KAAK,CAAC,IAAI,CAAC;aACX,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC;aAC7B,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAkB,CAAC,CAAC;IACtD,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,CAAC;IACZ,CAAC;AACH,CAAC"}