haansi 0.1.99 → 0.1.101
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/haansi.js +116 -15
- package/package.json +1 -1
package/dist/haansi.js
CHANGED
|
@@ -95,7 +95,7 @@ var require_package = __commonJS({
|
|
|
95
95
|
"package.json"(exports2, module2) {
|
|
96
96
|
module2.exports = {
|
|
97
97
|
name: "haansi",
|
|
98
|
-
version: "0.1.
|
|
98
|
+
version: "0.1.101",
|
|
99
99
|
description: "Haansi CLI - Session collector and MCP server for Claude Code & OpenAI Codex",
|
|
100
100
|
bin: {
|
|
101
101
|
haansi: "./dist/haansi.js"
|
|
@@ -86710,8 +86710,6 @@ function usageOf(output) {
|
|
|
86710
86710
|
}
|
|
86711
86711
|
function modelUnreachableOf(output, exitCode) {
|
|
86712
86712
|
if (exitCode === 0) return null;
|
|
86713
|
-
const used = usageOf(output);
|
|
86714
|
-
if ((used.outputTokens ?? 0) > 0) return null;
|
|
86715
86713
|
const objects = jsonLines(output);
|
|
86716
86714
|
const result = [...objects].reverse().find((o) => o.type === "result");
|
|
86717
86715
|
if (result && result.is_error === true) {
|
|
@@ -86719,6 +86717,8 @@ function modelUnreachableOf(output, exitCode) {
|
|
|
86719
86717
|
if (result.terminal_reason === "api_error" || MODEL_REFUSAL.test(said))
|
|
86720
86718
|
return clipReason(said || "the model's API returned an error");
|
|
86721
86719
|
}
|
|
86720
|
+
const used = usageOf(output);
|
|
86721
|
+
if ((used.outputTokens ?? 0) > 0) return null;
|
|
86722
86722
|
for (const o of [...objects].reverse()) {
|
|
86723
86723
|
const message = typeof o.message === "string" ? o.message : typeof o.error?.message === "string" ? String(o.error.message) : "";
|
|
86724
86724
|
if ((o.type === "error" || o.type === "turn.failed") && MODEL_REFUSAL.test(message))
|
|
@@ -86731,7 +86731,54 @@ function clipReason(text) {
|
|
|
86731
86731
|
const t = text.replace(/\s+/g, " ").trim();
|
|
86732
86732
|
return t.length > MAX_REASON_CHARS ? `${t.slice(0, MAX_REASON_CHARS - 1)}\u2026` : t;
|
|
86733
86733
|
}
|
|
86734
|
-
|
|
86734
|
+
function mcpServersOf(output) {
|
|
86735
|
+
const init2 = jsonLines(output).find(
|
|
86736
|
+
(o) => o.type === "system" && o.subtype === "init"
|
|
86737
|
+
);
|
|
86738
|
+
if (!init2 || !Array.isArray(init2.mcp_servers)) return null;
|
|
86739
|
+
return init2.mcp_servers.map((s) => s ?? {}).filter(
|
|
86740
|
+
(s) => typeof s.name === "string"
|
|
86741
|
+
).map((s) => ({
|
|
86742
|
+
name: s.name,
|
|
86743
|
+
status: typeof s.status === "string" ? s.status : "unknown"
|
|
86744
|
+
}));
|
|
86745
|
+
}
|
|
86746
|
+
function haansiServersDown(output) {
|
|
86747
|
+
const servers = mcpServersOf(output);
|
|
86748
|
+
if (!servers) return null;
|
|
86749
|
+
return servers.filter((s) => s.name.startsWith("haansi-") && s.status !== "connected").map((s) => `${s.name}: ${s.status}`);
|
|
86750
|
+
}
|
|
86751
|
+
function agentFailureLines(output, max2, scrub) {
|
|
86752
|
+
const out = [];
|
|
86753
|
+
for (const raw of output.trim().split(/\r?\n/)) {
|
|
86754
|
+
const line = raw.trim();
|
|
86755
|
+
if (!line) continue;
|
|
86756
|
+
if (line.startsWith("{") && line.endsWith("}")) {
|
|
86757
|
+
let event = null;
|
|
86758
|
+
try {
|
|
86759
|
+
event = JSON.parse(line);
|
|
86760
|
+
} catch {
|
|
86761
|
+
event = null;
|
|
86762
|
+
}
|
|
86763
|
+
if (event && typeof event === "object") {
|
|
86764
|
+
if (event.type !== "result") continue;
|
|
86765
|
+
const said = [
|
|
86766
|
+
typeof event.terminal_reason === "string" ? `terminal_reason=${event.terminal_reason}` : "",
|
|
86767
|
+
event.is_error === true ? "is_error" : "",
|
|
86768
|
+
typeof event.result === "string" ? event.result : ""
|
|
86769
|
+
].filter(Boolean).join(" ");
|
|
86770
|
+
if (said) out.push(said);
|
|
86771
|
+
continue;
|
|
86772
|
+
}
|
|
86773
|
+
}
|
|
86774
|
+
out.push(line);
|
|
86775
|
+
}
|
|
86776
|
+
return out.slice(-max2).map((l) => {
|
|
86777
|
+
const clean = scrub(l.replace(/\s+/g, " "));
|
|
86778
|
+
return clean.length > MAX_LOGGED_LINE_CHARS ? `${clean.slice(0, MAX_LOGGED_LINE_CHARS - 1)}\u2026` : clean;
|
|
86779
|
+
});
|
|
86780
|
+
}
|
|
86781
|
+
var NONE, count2, MODEL_REFUSAL, MAX_REASON_CHARS, MAX_LOGGED_LINE_CHARS;
|
|
86735
86782
|
var init_replay_usage = __esm({
|
|
86736
86783
|
"src/commands/replay-usage.ts"() {
|
|
86737
86784
|
"use strict";
|
|
@@ -86745,6 +86792,7 @@ var init_replay_usage = __esm({
|
|
|
86745
86792
|
count2 = (v) => typeof v === "number" && Number.isFinite(v) && v >= 0 ? v : null;
|
|
86746
86793
|
MODEL_REFUSAL = /credit balance is too low|insufficient[_ ](?:credit|quota|balance|funds)|quota (?:exceeded|exhausted)|exceeded your (?:current )?quota|invalid (?:x-)?api[_ -]?key|authentication[_ ]error|rate[_ ]limit(?:_error| exceeded| reached)|overloaded[_ ]error|api[_ ]error/i;
|
|
86747
86794
|
MAX_REASON_CHARS = 200;
|
|
86795
|
+
MAX_LOGGED_LINE_CHARS = 300;
|
|
86748
86796
|
}
|
|
86749
86797
|
});
|
|
86750
86798
|
|
|
@@ -88364,6 +88412,7 @@ function configFingerprintOf(c) {
|
|
|
88364
88412
|
return (0, import_node_crypto6.createHash)("sha256").update(
|
|
88365
88413
|
JSON.stringify({
|
|
88366
88414
|
agentCmd: c.agentCmd,
|
|
88415
|
+
judgeCmd: c.judgeCmd ?? null,
|
|
88367
88416
|
testCmd: c.testCmd,
|
|
88368
88417
|
setupCmd: c.setupCmd,
|
|
88369
88418
|
attempts: c.attempts,
|
|
@@ -88453,7 +88502,7 @@ async function runStoredReplays(options, deps) {
|
|
|
88453
88502
|
);
|
|
88454
88503
|
} else if (valid.length) {
|
|
88455
88504
|
const gate = await gateOnContext(deps, valid, {
|
|
88456
|
-
agentCmd: options.agentCmd,
|
|
88505
|
+
agentCmd: options.judgeCmd || options.agentCmd,
|
|
88457
88506
|
timeoutMs: options.timeoutMs,
|
|
88458
88507
|
concurrency: options.concurrency ?? 1,
|
|
88459
88508
|
limit: options.limit,
|
|
@@ -88580,7 +88629,7 @@ async function runStoredReplays(options, deps) {
|
|
|
88580
88629
|
...await judgeUse(deps, {
|
|
88581
88630
|
label,
|
|
88582
88631
|
taskId: task.taskId,
|
|
88583
|
-
agentCmd: options.agentCmd,
|
|
88632
|
+
agentCmd: options.judgeCmd || options.agentCmd,
|
|
88584
88633
|
timeoutMs: options.timeoutMs,
|
|
88585
88634
|
card: cardPrompt(task, null),
|
|
88586
88635
|
calls,
|
|
@@ -88610,7 +88659,7 @@ async function runStoredReplays(options, deps) {
|
|
|
88610
88659
|
const judged = await judgeRelevance(deps, {
|
|
88611
88660
|
label,
|
|
88612
88661
|
taskId: task.taskId,
|
|
88613
|
-
agentCmd: options.agentCmd,
|
|
88662
|
+
agentCmd: options.judgeCmd || options.agentCmd,
|
|
88614
88663
|
timeoutMs: options.timeoutMs,
|
|
88615
88664
|
card: cardPrompt(task, null),
|
|
88616
88665
|
targetChange: targetText,
|
|
@@ -88740,12 +88789,15 @@ var init_replay_tasks = __esm({
|
|
|
88740
88789
|
// src/commands/replay.ts
|
|
88741
88790
|
var replay_exports = {};
|
|
88742
88791
|
__export(replay_exports, {
|
|
88792
|
+
OUTPUT_HEAD_CHARS: () => OUTPUT_HEAD_CHARS,
|
|
88793
|
+
OUTPUT_TAIL_CHARS: () => OUTPUT_TAIL_CHARS,
|
|
88743
88794
|
SHA: () => SHA,
|
|
88744
88795
|
cleanClone: () => cleanClone,
|
|
88745
88796
|
dropLinks: () => dropLinks,
|
|
88746
88797
|
ensureCommit: () => ensureCommit,
|
|
88747
88798
|
githubRunRef: () => githubRunRef,
|
|
88748
88799
|
localDeps: () => localDeps,
|
|
88800
|
+
outputKeeper: () => outputKeeper,
|
|
88749
88801
|
parseReplayArgs: () => parseReplayArgs,
|
|
88750
88802
|
replay: () => replay,
|
|
88751
88803
|
replayArm: () => replayArm,
|
|
@@ -88788,6 +88840,7 @@ function parseReplayArgs(argv) {
|
|
|
88788
88840
|
"--family is retired: replays run on the tasks selected on Impact \u2192 Replays."
|
|
88789
88841
|
);
|
|
88790
88842
|
} else if (a === "--agent-cmd") options.agentCmd = next();
|
|
88843
|
+
else if (a === "--judge-cmd") options.judgeCmd = next() || null;
|
|
88791
88844
|
else if (a === "--test-cmd") options.testCmd = next();
|
|
88792
88845
|
else if (a === "--setup-cmd") options.setupCmd = next() || null;
|
|
88793
88846
|
else if (a === "--attempts") options.attempts = Number(next());
|
|
@@ -88971,7 +89024,11 @@ async function replayArm(deps, options, task, arm, promptOf, inspect, agentEnv =
|
|
|
88971
89024
|
});
|
|
88972
89025
|
if (agent.code !== 0) {
|
|
88973
89026
|
deps.log(`\u2026 ${label}: the agent exited with code ${agent.code}`);
|
|
88974
|
-
for (const line of
|
|
89027
|
+
for (const line of agentFailureLines(
|
|
89028
|
+
agent.output,
|
|
89029
|
+
FAILURE_LINES2,
|
|
89030
|
+
scrubSecrets
|
|
89031
|
+
))
|
|
88975
89032
|
deps.log(` ${line}`);
|
|
88976
89033
|
}
|
|
88977
89034
|
const unreachable2 = modelUnreachableOf(agent.output, agent.code);
|
|
@@ -88979,6 +89036,17 @@ async function replayArm(deps, options, task, arm, promptOf, inspect, agentEnv =
|
|
|
88979
89036
|
throw new Error(
|
|
88980
89037
|
`the agent could not reach its model (${unreachable2}); not graded, run it again`
|
|
88981
89038
|
);
|
|
89039
|
+
if (arm === "haansi") {
|
|
89040
|
+
const down = haansiServersDown(agent.output);
|
|
89041
|
+
if (down === null)
|
|
89042
|
+
deps.log(
|
|
89043
|
+
`\u2026 ${label}: the agent did not report its MCP servers (use --output-format stream-json --verbose to check them)`
|
|
89044
|
+
);
|
|
89045
|
+
else if (down.length)
|
|
89046
|
+
throw new Error(
|
|
89047
|
+
`Haansi's servers did not load for the agent (${down.join(", ")}); not graded, run it again`
|
|
89048
|
+
);
|
|
89049
|
+
}
|
|
88982
89050
|
const gradeSha = task.testNames?.length ? task.baseSha : task.targetSha;
|
|
88983
89051
|
const own = new Map(files.map((f) => [f, deps.readFile((0, import_node_path16.join)(dir, f))]));
|
|
88984
89052
|
for (const f of files) {
|
|
@@ -89079,6 +89147,31 @@ async function replayArm(deps, options, task, arm, promptOf, inspect, agentEnv =
|
|
|
89079
89147
|
if (!agentEnv.HAANSI_REPLAY_MCP_CONFIG) deps.removeFile(`${dir}.mcp.json`);
|
|
89080
89148
|
}
|
|
89081
89149
|
}
|
|
89150
|
+
function outputKeeper(headChars = OUTPUT_HEAD_CHARS, tailChars = OUTPUT_TAIL_CHARS) {
|
|
89151
|
+
let head = "";
|
|
89152
|
+
let tail = "";
|
|
89153
|
+
let dropped = false;
|
|
89154
|
+
return {
|
|
89155
|
+
add(text) {
|
|
89156
|
+
if (head.length < headChars) {
|
|
89157
|
+
const room = headChars - head.length;
|
|
89158
|
+
head += text.slice(0, room);
|
|
89159
|
+
text = text.slice(room);
|
|
89160
|
+
}
|
|
89161
|
+
if (!text) return;
|
|
89162
|
+
tail += text;
|
|
89163
|
+
if (tail.length > tailChars) {
|
|
89164
|
+
tail = tail.slice(-tailChars);
|
|
89165
|
+
dropped = true;
|
|
89166
|
+
}
|
|
89167
|
+
},
|
|
89168
|
+
text() {
|
|
89169
|
+
return dropped ? `${head}
|
|
89170
|
+
\u2026 (output trimmed)
|
|
89171
|
+
${tail}` : head + tail;
|
|
89172
|
+
}
|
|
89173
|
+
};
|
|
89174
|
+
}
|
|
89082
89175
|
function runCommand(file2, args, cwd, env, timeoutMs, stdoutOnly = false) {
|
|
89083
89176
|
return new Promise((resolvePromise) => {
|
|
89084
89177
|
const child = (0, import_node_child_process5.spawn)(file2, args, {
|
|
@@ -89087,9 +89180,9 @@ function runCommand(file2, args, cwd, env, timeoutMs, stdoutOnly = false) {
|
|
|
89087
89180
|
stdio: ["ignore", "pipe", "pipe"]
|
|
89088
89181
|
});
|
|
89089
89182
|
let output = "";
|
|
89090
|
-
const
|
|
89091
|
-
|
|
89092
|
-
|
|
89183
|
+
const kept = outputKeeper();
|
|
89184
|
+
const keep = (chunk) => kept.add(chunk.toString());
|
|
89185
|
+
const finalOutput = () => stdoutOnly ? output : kept.text();
|
|
89093
89186
|
if (stdoutOnly) {
|
|
89094
89187
|
const chunks = [];
|
|
89095
89188
|
child.stdout.on("data", (chunk) => chunks.push(chunk));
|
|
@@ -89109,13 +89202,13 @@ function runCommand(file2, args, cwd, env, timeoutMs, stdoutOnly = false) {
|
|
|
89109
89202
|
clearTimeout(timer);
|
|
89110
89203
|
resolvePromise({
|
|
89111
89204
|
code: code ?? 1,
|
|
89112
|
-
output,
|
|
89205
|
+
output: finalOutput(),
|
|
89113
89206
|
...timedOut ? { timedOut } : {}
|
|
89114
89207
|
});
|
|
89115
89208
|
});
|
|
89116
89209
|
child.on("error", () => {
|
|
89117
89210
|
clearTimeout(timer);
|
|
89118
|
-
resolvePromise({ code: 1, output });
|
|
89211
|
+
resolvePromise({ code: 1, output: finalOutput() });
|
|
89119
89212
|
});
|
|
89120
89213
|
});
|
|
89121
89214
|
}
|
|
@@ -89297,7 +89390,7 @@ Baseline: ${arm("baseline")} passed \xB7 With Haansi: ${arm("haansi")} passed \x
|
|
|
89297
89390
|
(0, import_node_fs13.rmSync)(work, { recursive: true, force: true });
|
|
89298
89391
|
}
|
|
89299
89392
|
}
|
|
89300
|
-
var import_node_child_process5, import_node_fs13, import_node_os11, import_node_path16, SHA, ARMS, FEEDBACK_CHARS, FAILURE_LINES2, REPLAY_BRANCH, USAGE2;
|
|
89393
|
+
var import_node_child_process5, import_node_fs13, import_node_os11, import_node_path16, SHA, ARMS, FEEDBACK_CHARS, FAILURE_LINES2, REPLAY_BRANCH, OUTPUT_HEAD_CHARS, OUTPUT_TAIL_CHARS, USAGE2;
|
|
89301
89394
|
var init_replay = __esm({
|
|
89302
89395
|
"src/commands/replay.ts"() {
|
|
89303
89396
|
"use strict";
|
|
@@ -89319,9 +89412,12 @@ var init_replay = __esm({
|
|
|
89319
89412
|
FEEDBACK_CHARS = 6e3;
|
|
89320
89413
|
FAILURE_LINES2 = 8;
|
|
89321
89414
|
REPLAY_BRANCH = "replay";
|
|
89415
|
+
OUTPUT_HEAD_CHARS = 4e4;
|
|
89416
|
+
OUTPUT_TAIL_CHARS = 2e5;
|
|
89322
89417
|
USAGE2 = `Usage: haansi replay run --agent-cmd "<cmd>" --test-cmd "<cmd>"
|
|
89323
89418
|
[--setup-cmd "<cmd>"] [--attempts 3] [--arms baseline,haansi] [--limit n]
|
|
89324
89419
|
[--timeout-min 20] [--agent-name <name>] [--model <name>]
|
|
89420
|
+
[--judge-cmd "<cmd>"]
|
|
89325
89421
|
[--run-ref <url>] [--dry-run] [--repo-commands] [--no-diffs]
|
|
89326
89422
|
[--shard i/n] [--concurrency k] [--force] [--max-hours h] [--any-context]
|
|
89327
89423
|
[--setup-cache <files> --setup-cache-path <paths> [--setup-cache-dir <dir>]]
|
|
@@ -89356,6 +89452,11 @@ A task with nothing relevant Haansi can return is skipped, and checked again
|
|
|
89356
89452
|
once the organization gains new Haansi content. --any-context replays every
|
|
89357
89453
|
valid task without the check (diagnostics).
|
|
89358
89454
|
|
|
89455
|
+
--judge-cmd runs the judges (whether the agent used what Haansi returned,
|
|
89456
|
+
which of it was relevant, and the context check before replaying) with their
|
|
89457
|
+
own command, e.g. a stronger model than a cheap agent's; --agent-cmd when not
|
|
89458
|
+
given.
|
|
89459
|
+
|
|
89359
89460
|
--repo-commands uses the setup and test commands an org owner or admin saved for the
|
|
89360
89461
|
repository in Haansi (Impact \u2192 Replays) for any not given here.
|
|
89361
89462
|
|
|
@@ -89368,7 +89469,7 @@ Run from a clone of the repository. Example (GitHub Actions):
|
|
|
89368
89469
|
haansi replay run --setup-cmd 'npm ci' \\
|
|
89369
89470
|
--container node:22 --container-env ANTHROPIC_API_KEY \\
|
|
89370
89471
|
--container-setup 'npm install -g haansi @anthropic-ai/claude-code' \\
|
|
89371
|
-
--agent-cmd 'claude -p "$HAANSI_REPLAY_PROMPT" --output-format json --permission-mode acceptEdits --mcp-config "$HAANSI_REPLAY_MCP_CONFIG" --allowedTools mcp__haansi-solutions,mcp__haansi-exchange,mcp__haansi-digests,mcp__haansi-onboarding,mcp__haansi-skills' \\
|
|
89472
|
+
--agent-cmd 'claude -p "$HAANSI_REPLAY_PROMPT" --output-format stream-json --verbose --permission-mode acceptEdits --mcp-config "$HAANSI_REPLAY_MCP_CONFIG" --allowedTools mcp__haansi-solutions,mcp__haansi-exchange,mcp__haansi-digests,mcp__haansi-onboarding,mcp__haansi-skills' \\
|
|
89372
89473
|
--test-cmd 'npx vitest run $HAANSI_REPLAY_TEST_FILES -t "$HAANSI_REPLAY_TEST_PATTERN"' \\
|
|
89373
89474
|
--agent-name claude-code --model claude-sonnet-5-5`;
|
|
89374
89475
|
}
|