haansi 0.1.104 → 0.1.106
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 +285 -96
- 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: "0.1.106",
|
|
99
99
|
description: "Haansi CLI - Session collector and MCP server for Claude Code & OpenAI Codex",
|
|
100
100
|
bin: {
|
|
101
101
|
haansi: "./dist/haansi.js"
|
|
@@ -547,10 +547,10 @@ __export(setup_mcp_exports, {
|
|
|
547
547
|
});
|
|
548
548
|
function parseTarget(arg) {
|
|
549
549
|
if (!arg || arg === "all") return "all";
|
|
550
|
-
if (arg === "solutions" || arg === "exchange" || arg === "digests" || arg === "onboarding" || arg === "skills")
|
|
550
|
+
if (arg === "solutions" || arg === "exchange" || arg === "digests" || arg === "onboarding" || arg === "skills" || arg === "incidents")
|
|
551
551
|
return arg;
|
|
552
552
|
throw new Error(
|
|
553
|
-
`Unknown target: ${arg}. Use one of: solutions, exchange, digests, onboarding, skills, all (default).`
|
|
553
|
+
`Unknown target: ${arg}. Use one of: solutions, exchange, digests, onboarding, skills, incidents, all (default).`
|
|
554
554
|
);
|
|
555
555
|
}
|
|
556
556
|
function httpServerUrl(path2) {
|
|
@@ -792,6 +792,13 @@ var init_setup_mcp = __esm({
|
|
|
792
792
|
target: "skills",
|
|
793
793
|
name: "haansi-skills",
|
|
794
794
|
path: "/api/v1/memory/skills/mcp"
|
|
795
|
+
},
|
|
796
|
+
// Past incidents: what happened, how it was resolved, and what resembles
|
|
797
|
+
// the problem at hand (served as of a replay's start in a replay).
|
|
798
|
+
{
|
|
799
|
+
target: "incidents",
|
|
800
|
+
name: "haansi-incidents",
|
|
801
|
+
path: "/api/v1/memory/processes/incidents/mcp"
|
|
795
802
|
}
|
|
796
803
|
];
|
|
797
804
|
HSK_BEARER = /Bearer hsk_[A-Za-z0-9_-]+/;
|
|
@@ -60235,6 +60242,38 @@ var init_search_solutions_tool = __esm({
|
|
|
60235
60242
|
}
|
|
60236
60243
|
});
|
|
60237
60244
|
|
|
60245
|
+
// ../service-capture/agent-sessions/get-solution-tool.ts
|
|
60246
|
+
function getSolutionTool(contextSchema2) {
|
|
60247
|
+
return {
|
|
60248
|
+
name: "get_solution",
|
|
60249
|
+
description: "Read one solution from your organization's knowledge base in full: the problem, the solution, every step with why it was needed, and the files it changed. Pass the id a search_solutions result names (solution:<id>). Search first; open the results that look like your problem.",
|
|
60250
|
+
inputSchema: {
|
|
60251
|
+
type: "object",
|
|
60252
|
+
properties: {
|
|
60253
|
+
id: {
|
|
60254
|
+
type: "string",
|
|
60255
|
+
description: "The solution's id, as a search result names it"
|
|
60256
|
+
},
|
|
60257
|
+
_context: contextSchema2
|
|
60258
|
+
},
|
|
60259
|
+
required: ["id"]
|
|
60260
|
+
}
|
|
60261
|
+
};
|
|
60262
|
+
}
|
|
60263
|
+
function getSolutionPath(id) {
|
|
60264
|
+
const match = typeof id === "string" ? ID.exec(id.trim()) : null;
|
|
60265
|
+
return match ? `/memory/solutions/${match[1].toLowerCase()}` : null;
|
|
60266
|
+
}
|
|
60267
|
+
var ID, NOT_A_SOLUTION_ID, NO_SUCH_SOLUTION;
|
|
60268
|
+
var init_get_solution_tool = __esm({
|
|
60269
|
+
"../service-capture/agent-sessions/get-solution-tool.ts"() {
|
|
60270
|
+
"use strict";
|
|
60271
|
+
ID = /^(?:solution:)?([0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12})$/i;
|
|
60272
|
+
NOT_A_SOLUTION_ID = "Pass the id a search_solutions result names, e.g. solution:1b2c3d4e-\u2026.";
|
|
60273
|
+
NO_SUCH_SOLUTION = "No such solution: it may have been removed, or it is not one you can read.";
|
|
60274
|
+
}
|
|
60275
|
+
});
|
|
60276
|
+
|
|
60238
60277
|
// ../../node_modules/depd/index.js
|
|
60239
60278
|
var require_depd = __commonJS({
|
|
60240
60279
|
"../../node_modules/depd/index.js"(exports2, module2) {
|
|
@@ -84401,6 +84440,7 @@ var init_mcp_server2 = __esm({
|
|
|
84401
84440
|
init_stdio2();
|
|
84402
84441
|
init_entrypoint();
|
|
84403
84442
|
init_search_solutions_tool();
|
|
84443
|
+
init_get_solution_tool();
|
|
84404
84444
|
init_types4();
|
|
84405
84445
|
(0, import_dotenv3.config)({ path: (0, import_path5.resolve)(__dirname, "../../../.env") });
|
|
84406
84446
|
DEFAULT_API_URL3 = SERVICE_URL_DEFAULTS.haansiApiBaseUrl;
|
|
@@ -84438,6 +84478,7 @@ var init_mcp_server2 = __esm({
|
|
|
84438
84478
|
server.setRequestHandler(ListToolsRequestSchema, async () => ({
|
|
84439
84479
|
tools: [
|
|
84440
84480
|
searchSolutionsTool(contextSchema),
|
|
84481
|
+
getSolutionTool(contextSchema),
|
|
84441
84482
|
{
|
|
84442
84483
|
name: "list_recent_solutions",
|
|
84443
84484
|
description: "List the most recently added distilled paths from the knowledge base.",
|
|
@@ -84577,6 +84618,34 @@ var init_mcp_server2 = __esm({
|
|
|
84577
84618
|
]
|
|
84578
84619
|
};
|
|
84579
84620
|
}
|
|
84621
|
+
case "get_solution": {
|
|
84622
|
+
const { id, _context } = args;
|
|
84623
|
+
const path2 = getSolutionPath(id);
|
|
84624
|
+
if (!path2) {
|
|
84625
|
+
return {
|
|
84626
|
+
content: [{ type: "text", text: NOT_A_SOLUTION_ID }],
|
|
84627
|
+
isError: true
|
|
84628
|
+
};
|
|
84629
|
+
}
|
|
84630
|
+
const extraHeaders = {};
|
|
84631
|
+
if (_context?.userId) {
|
|
84632
|
+
extraHeaders["X-Session-User-Id"] = String(_context.userId);
|
|
84633
|
+
}
|
|
84634
|
+
if (_context?.orgId) {
|
|
84635
|
+
extraHeaders["X-Session-Org-Id"] = String(_context.orgId);
|
|
84636
|
+
}
|
|
84637
|
+
try {
|
|
84638
|
+
const data = await apiGet2(path2, { extraHeaders });
|
|
84639
|
+
return {
|
|
84640
|
+
content: [{ type: "text", text: data.text ?? NO_SUCH_SOLUTION }]
|
|
84641
|
+
};
|
|
84642
|
+
} catch (err2) {
|
|
84643
|
+
if (err2 instanceof ApiError && err2.status === 404) {
|
|
84644
|
+
return { content: [{ type: "text", text: NO_SUCH_SOLUTION }] };
|
|
84645
|
+
}
|
|
84646
|
+
throw err2;
|
|
84647
|
+
}
|
|
84648
|
+
}
|
|
84580
84649
|
case "list_recent_solutions": {
|
|
84581
84650
|
const { limit = 10, _context } = args;
|
|
84582
84651
|
const n = Math.min(Math.max(1, limit), 50);
|
|
@@ -86485,14 +86554,14 @@ var init_replay_focus = __esm({
|
|
|
86485
86554
|
|
|
86486
86555
|
// src/commands/replay-preflight.ts
|
|
86487
86556
|
function refusalOf(status, body) {
|
|
86488
|
-
let
|
|
86557
|
+
let said2 = body.trim();
|
|
86489
86558
|
try {
|
|
86490
86559
|
const parsed = JSON.parse(body);
|
|
86491
86560
|
const m = parsed.message ?? parsed.error;
|
|
86492
|
-
if (typeof m === "string" && m.trim())
|
|
86561
|
+
if (typeof m === "string" && m.trim()) said2 = m.trim();
|
|
86493
86562
|
} catch {
|
|
86494
86563
|
}
|
|
86495
|
-
const short =
|
|
86564
|
+
const short = said2.replace(/\s+/g, " ").slice(0, 120);
|
|
86496
86565
|
return short ? `${status}: ${short}` : String(status);
|
|
86497
86566
|
}
|
|
86498
86567
|
async function check2(fetchImpl, name, url2, init2) {
|
|
@@ -86713,6 +86782,18 @@ var init_replay_preflight = __esm({
|
|
|
86713
86782
|
});
|
|
86714
86783
|
|
|
86715
86784
|
// src/commands/replay-fallback.ts
|
|
86785
|
+
function solutionsSearchQuery(q, hints) {
|
|
86786
|
+
const params = new URLSearchParams();
|
|
86787
|
+
params.set("q", q);
|
|
86788
|
+
params.set("top_k", String(SOLUTIONS_PER_QUERY));
|
|
86789
|
+
for (const t of clipList2(hints?.failingTests))
|
|
86790
|
+
params.append("failing_tests", t);
|
|
86791
|
+
const error2 = hints?.error?.trim();
|
|
86792
|
+
if (error2) params.set("error", error2.slice(0, MAX_HINT_ERROR_CHARS2));
|
|
86793
|
+
for (const f of clipList2(hints?.files)) params.append("files", f);
|
|
86794
|
+
for (const n of clipList2(hints?.names)) params.append("names", n);
|
|
86795
|
+
return params.toString();
|
|
86796
|
+
}
|
|
86716
86797
|
async function searchHaansi(session, ask, fetchImpl = fetch) {
|
|
86717
86798
|
const base = session.apiUrl.replace(/\/+$/, "");
|
|
86718
86799
|
const headers = {
|
|
@@ -86721,22 +86802,20 @@ async function searchHaansi(session, ask, fetchImpl = fetch) {
|
|
|
86721
86802
|
};
|
|
86722
86803
|
const solutions = ask.queries.map(
|
|
86723
86804
|
(q) => fetchImpl(
|
|
86724
|
-
`${base}/api/v1/memory/solutions/search
|
|
86805
|
+
`${base}/api/v1/memory/solutions/search?${solutionsSearchQuery(q, ask.hints)}`,
|
|
86725
86806
|
{ method: "GET", headers }
|
|
86726
86807
|
)
|
|
86727
86808
|
);
|
|
86728
|
-
const
|
|
86729
|
-
|
|
86730
|
-
|
|
86731
|
-
const
|
|
86732
|
-
...headers,
|
|
86733
|
-
"Content-Type": "application/json",
|
|
86734
|
-
Accept: "application/json, text/event-stream"
|
|
86735
|
-
};
|
|
86809
|
+
const mcp = async (server2, calls) => {
|
|
86810
|
+
const path2 = HTTP_SERVERS.find((s) => s.name === server2)?.path;
|
|
86811
|
+
if (!path2 || calls.length === 0) return;
|
|
86812
|
+
const url2 = `${base}${path2}`;
|
|
86736
86813
|
const rpc = (id, method, params, sid2) => fetchImpl(url2, {
|
|
86737
86814
|
method: "POST",
|
|
86738
86815
|
headers: {
|
|
86739
|
-
...
|
|
86816
|
+
...headers,
|
|
86817
|
+
"Content-Type": "application/json",
|
|
86818
|
+
Accept: "application/json, text/event-stream",
|
|
86740
86819
|
...sid2 ? { "Mcp-Session-Id": sid2 } : {}
|
|
86741
86820
|
},
|
|
86742
86821
|
body: JSON.stringify({ jsonrpc: "2.0", id, method, params })
|
|
@@ -86748,57 +86827,54 @@ async function searchHaansi(session, ask, fetchImpl = fetch) {
|
|
|
86748
86827
|
});
|
|
86749
86828
|
if (!init2.ok) return;
|
|
86750
86829
|
const sid = init2.headers?.get("mcp-session-id") ?? void 0;
|
|
86751
|
-
await Promise.allSettled(
|
|
86752
|
-
|
|
86753
|
-
2,
|
|
86754
|
-
|
|
86755
|
-
|
|
86756
|
-
|
|
86757
|
-
|
|
86758
|
-
|
|
86759
|
-
|
|
86760
|
-
|
|
86761
|
-
|
|
86762
|
-
|
|
86763
|
-
|
|
86764
|
-
...ask.
|
|
86765
|
-
|
|
86766
|
-
|
|
86767
|
-
|
|
86768
|
-
|
|
86769
|
-
|
|
86770
|
-
|
|
86771
|
-
|
|
86772
|
-
|
|
86773
|
-
|
|
86774
|
-
|
|
86775
|
-
|
|
86776
|
-
|
|
86777
|
-
|
|
86778
|
-
|
|
86779
|
-
|
|
86780
|
-
|
|
86781
|
-
|
|
86782
|
-
|
|
86783
|
-
|
|
86784
|
-
|
|
86785
|
-
}
|
|
86786
|
-
},
|
|
86787
|
-
sid
|
|
86788
|
-
).then((r) => r.text().then(rpcResult))
|
|
86789
|
-
] : []
|
|
86790
|
-
]);
|
|
86791
|
-
})() : Promise.resolve();
|
|
86792
|
-
await Promise.allSettled([...solutions, skills2]);
|
|
86830
|
+
await Promise.allSettled(
|
|
86831
|
+
calls.map(
|
|
86832
|
+
(call, i) => rpc(i + 2, "tools/call", call, sid).then(
|
|
86833
|
+
(r) => r.text().then(rpcResult)
|
|
86834
|
+
)
|
|
86835
|
+
)
|
|
86836
|
+
);
|
|
86837
|
+
};
|
|
86838
|
+
const first = ask.queries[0];
|
|
86839
|
+
const searchWith = (name) => first ? [{ name, arguments: { query: first, limit: OTHER_SOURCE_LIMIT } }] : [];
|
|
86840
|
+
const skills2 = mcp("haansi-skills", [
|
|
86841
|
+
{
|
|
86842
|
+
name: "match_playbook",
|
|
86843
|
+
arguments: { task: ask.task, ...ask.error ? { error: ask.error } : {} }
|
|
86844
|
+
},
|
|
86845
|
+
...first ? [{ name: "search_skills", arguments: { query: first, limit: 5 } }] : [],
|
|
86846
|
+
...ask.paths?.length ? [
|
|
86847
|
+
{
|
|
86848
|
+
name: "get_prior_attempts",
|
|
86849
|
+
arguments: {
|
|
86850
|
+
paths: ask.paths,
|
|
86851
|
+
...first ? { query: first } : {},
|
|
86852
|
+
limit: PRIOR_ATTEMPTS_LIMIT
|
|
86853
|
+
}
|
|
86854
|
+
}
|
|
86855
|
+
] : []
|
|
86856
|
+
]);
|
|
86857
|
+
const others = [
|
|
86858
|
+
mcp("haansi-exchange", searchWith("search_exchange")),
|
|
86859
|
+
mcp("haansi-digests", searchWith("search_digests")),
|
|
86860
|
+
mcp("haansi-onboarding", searchWith("search_guides")),
|
|
86861
|
+
mcp("haansi-incidents", searchWith("search_prior_incidents"))
|
|
86862
|
+
];
|
|
86863
|
+
await Promise.allSettled([...solutions, skills2, ...others]);
|
|
86793
86864
|
}
|
|
86794
|
-
var SOLUTIONS_PER_QUERY, PRIOR_ATTEMPTS_LIMIT;
|
|
86865
|
+
var SOLUTIONS_PER_QUERY, MAX_HINT_ITEMS2, MAX_HINT_ITEM_CHARS2, MAX_HINT_ERROR_CHARS2, clipList2, PRIOR_ATTEMPTS_LIMIT, OTHER_SOURCE_LIMIT;
|
|
86795
86866
|
var init_replay_fallback = __esm({
|
|
86796
86867
|
"src/commands/replay-fallback.ts"() {
|
|
86797
86868
|
"use strict";
|
|
86798
86869
|
init_setup_mcp();
|
|
86799
86870
|
init_replay_preflight();
|
|
86800
86871
|
SOLUTIONS_PER_QUERY = 5;
|
|
86872
|
+
MAX_HINT_ITEMS2 = 10;
|
|
86873
|
+
MAX_HINT_ITEM_CHARS2 = 300;
|
|
86874
|
+
MAX_HINT_ERROR_CHARS2 = 1e3;
|
|
86875
|
+
clipList2 = (items) => (items ?? []).map((v) => v.trim().slice(0, MAX_HINT_ITEM_CHARS2)).filter(Boolean).slice(0, MAX_HINT_ITEMS2);
|
|
86801
86876
|
PRIOR_ATTEMPTS_LIMIT = 3;
|
|
86877
|
+
OTHER_SOURCE_LIMIT = 5;
|
|
86802
86878
|
}
|
|
86803
86879
|
});
|
|
86804
86880
|
|
|
@@ -86858,9 +86934,9 @@ function modelUnreachableOf(output, exitCode) {
|
|
|
86858
86934
|
const objects = jsonLines(output);
|
|
86859
86935
|
const result = [...objects].reverse().find((o) => o.type === "result");
|
|
86860
86936
|
if (result && result.is_error === true) {
|
|
86861
|
-
const
|
|
86862
|
-
if (result.terminal_reason === "api_error" || MODEL_REFUSAL.test(
|
|
86863
|
-
return clipReason(
|
|
86937
|
+
const said2 = typeof result.result === "string" ? result.result : "";
|
|
86938
|
+
if (result.terminal_reason === "api_error" || MODEL_REFUSAL.test(said2))
|
|
86939
|
+
return clipReason(said2 || "the model's API returned an error");
|
|
86864
86940
|
}
|
|
86865
86941
|
const used = usageOf(output);
|
|
86866
86942
|
if ((used.outputTokens ?? 0) > 0) return null;
|
|
@@ -86907,12 +86983,12 @@ function agentFailureLines(output, max2, scrub) {
|
|
|
86907
86983
|
}
|
|
86908
86984
|
if (event && typeof event === "object") {
|
|
86909
86985
|
if (event.type !== "result") continue;
|
|
86910
|
-
const
|
|
86986
|
+
const said2 = [
|
|
86911
86987
|
typeof event.terminal_reason === "string" ? `terminal_reason=${event.terminal_reason}` : "",
|
|
86912
86988
|
event.is_error === true ? "is_error" : "",
|
|
86913
86989
|
typeof event.result === "string" ? event.result : ""
|
|
86914
86990
|
].filter(Boolean).join(" ");
|
|
86915
|
-
if (
|
|
86991
|
+
if (said2) out.push(said2);
|
|
86916
86992
|
continue;
|
|
86917
86993
|
}
|
|
86918
86994
|
}
|
|
@@ -87954,13 +88030,20 @@ function parseRelevance(text, shown) {
|
|
|
87954
88030
|
return null;
|
|
87955
88031
|
}
|
|
87956
88032
|
if (!Array.isArray(v.items)) return null;
|
|
87957
|
-
const
|
|
88033
|
+
const onPoint = /* @__PURE__ */ new Set();
|
|
88034
|
+
const nearby = /* @__PURE__ */ new Set();
|
|
87958
88035
|
for (const item of v.items) {
|
|
87959
88036
|
const it = item ?? {};
|
|
87960
88037
|
if (typeof it.id !== "string" || !shown.has(it.id)) continue;
|
|
87961
|
-
if (it.
|
|
88038
|
+
if (it.grade === "on_point") {
|
|
88039
|
+
const e = it.evidence ?? {};
|
|
88040
|
+
if (said(e.quote) && said(e.change)) onPoint.add(it.id);
|
|
88041
|
+
else nearby.add(it.id);
|
|
88042
|
+
} else if (it.grade === "nearby") nearby.add(it.id);
|
|
88043
|
+
else if (it.grade === void 0 && it.relevant === true) onPoint.add(it.id);
|
|
87962
88044
|
}
|
|
87963
|
-
|
|
88045
|
+
for (const id of onPoint) nearby.delete(id);
|
|
88046
|
+
return { onPoint: [...onPoint].sort(), nearby: [...nearby].sort() };
|
|
87964
88047
|
}
|
|
87965
88048
|
function knowledgeOf(relevantIds, reachedAgent) {
|
|
87966
88049
|
if (relevantIds.some((id) => reachedAgent.has(id))) return "returned";
|
|
@@ -87986,6 +88069,7 @@ async function judgeRelevance(deps, input) {
|
|
|
87986
88069
|
return {
|
|
87987
88070
|
relevantIds: [],
|
|
87988
88071
|
recallIds: [],
|
|
88072
|
+
nearbyIds: [],
|
|
87989
88073
|
knowledge: "none",
|
|
87990
88074
|
judgeTag: null
|
|
87991
88075
|
};
|
|
@@ -87993,6 +88077,7 @@ async function judgeRelevance(deps, input) {
|
|
|
87993
88077
|
return {
|
|
87994
88078
|
relevantIds: null,
|
|
87995
88079
|
recallIds: null,
|
|
88080
|
+
nearbyIds: null,
|
|
87996
88081
|
knowledge: null,
|
|
87997
88082
|
judgeTag: null
|
|
87998
88083
|
};
|
|
@@ -88028,21 +88113,24 @@ async function judgeRelevance(deps, input) {
|
|
|
88028
88113
|
},
|
|
88029
88114
|
input.timeoutMs
|
|
88030
88115
|
);
|
|
88031
|
-
const
|
|
88032
|
-
if (!
|
|
88116
|
+
const grades = parseRelevance(deps.readFile(verdictFile) ?? "", shown) ?? parseRelevance(ran.output, shown);
|
|
88117
|
+
if (!grades) {
|
|
88033
88118
|
deps.log(
|
|
88034
88119
|
`\u2026 ${input.label} [haansi]: the relevance judge gave no verdict`
|
|
88035
88120
|
);
|
|
88036
88121
|
return {
|
|
88037
88122
|
relevantIds: null,
|
|
88038
88123
|
recallIds: null,
|
|
88124
|
+
nearbyIds: null,
|
|
88039
88125
|
knowledge: null,
|
|
88040
88126
|
judgeTag: tag
|
|
88041
88127
|
};
|
|
88042
88128
|
}
|
|
88129
|
+
const relevant = grades.onPoint;
|
|
88043
88130
|
return {
|
|
88044
88131
|
relevantIds: relevant,
|
|
88045
88132
|
recallIds: relevant.filter((id) => !reachedIds.has(id)),
|
|
88133
|
+
nearbyIds: grades.nearby,
|
|
88046
88134
|
knowledge: knowledgeOf(relevant, reachedIds),
|
|
88047
88135
|
judgeTag: tag
|
|
88048
88136
|
};
|
|
@@ -88050,7 +88138,7 @@ async function judgeRelevance(deps, input) {
|
|
|
88050
88138
|
deps.removeDir(dir);
|
|
88051
88139
|
}
|
|
88052
88140
|
}
|
|
88053
|
-
var import_node_path14, MAX_CHANGE_CHARS2, MAX_ITEMS_CHARS, QUERY_CHARS, TERMS, clip2, COMMON;
|
|
88141
|
+
var import_node_path14, MAX_CHANGE_CHARS2, MAX_ITEMS_CHARS, QUERY_CHARS, TERMS, clip2, COMMON, said;
|
|
88054
88142
|
var init_replay_relevance = __esm({
|
|
88055
88143
|
"src/commands/replay-relevance.ts"() {
|
|
88056
88144
|
"use strict";
|
|
@@ -88065,6 +88153,7 @@ var init_replay_relevance = __esm({
|
|
|
88065
88153
|
COMMON = new Set(
|
|
88066
88154
|
"this that with from have will when then than they them their there what which while where return const function export import class true false null undefined async await string number value values type types test tests expect should data item items list name names error errors".split(" ")
|
|
88067
88155
|
);
|
|
88156
|
+
said = (v) => typeof v === "string" && v.trim() !== "";
|
|
88068
88157
|
}
|
|
88069
88158
|
});
|
|
88070
88159
|
|
|
@@ -89231,6 +89320,17 @@ var init_replay_spec = __esm({
|
|
|
89231
89320
|
});
|
|
89232
89321
|
|
|
89233
89322
|
// src/commands/replay-context.ts
|
|
89323
|
+
function storedScoreOf(task) {
|
|
89324
|
+
return {
|
|
89325
|
+
onPoint: task.context?.onPoint ?? 0,
|
|
89326
|
+
nearby: task.context?.nearby ?? 0
|
|
89327
|
+
};
|
|
89328
|
+
}
|
|
89329
|
+
function rankFound(tasks, scoreOf) {
|
|
89330
|
+
return tasks.map((task) => ({ task, score: scoreOf(task) })).sort(
|
|
89331
|
+
(a, b) => b.score.onPoint - a.score.onPoint || b.score.nearby - a.score.nearby || a.task.place - b.task.place
|
|
89332
|
+
).map((r) => r.task);
|
|
89333
|
+
}
|
|
89234
89334
|
function stateOf(knowledge) {
|
|
89235
89335
|
if (knowledge === "returned") return "found";
|
|
89236
89336
|
return knowledge;
|
|
@@ -89261,6 +89361,7 @@ function visibleAsk(task) {
|
|
|
89261
89361
|
].join("\n");
|
|
89262
89362
|
let paths;
|
|
89263
89363
|
let queries;
|
|
89364
|
+
let hints;
|
|
89264
89365
|
if (failing?.tests.length) {
|
|
89265
89366
|
const named = failing.tests.map(fileOfTest).filter((f) => !!f);
|
|
89266
89367
|
paths = [
|
|
@@ -89270,11 +89371,17 @@ function visibleAsk(task) {
|
|
|
89270
89371
|
])
|
|
89271
89372
|
].slice(0, MAX_FAILING_PATHS);
|
|
89272
89373
|
const files = [...new Set(paths.map(baseNameOf))];
|
|
89374
|
+
const errorLine = errorLineOf(failing.output);
|
|
89273
89375
|
queries = [
|
|
89274
89376
|
...failing.tests.slice(0, FAILING_QUERIES).map(titleOfTest),
|
|
89275
|
-
|
|
89377
|
+
errorLine ?? "",
|
|
89276
89378
|
files.slice(0, 6).join(" ")
|
|
89277
89379
|
];
|
|
89380
|
+
hints = {
|
|
89381
|
+
failingTests: failing.tests,
|
|
89382
|
+
...errorLine ? { error: errorLine } : {},
|
|
89383
|
+
files: paths
|
|
89384
|
+
};
|
|
89278
89385
|
} else {
|
|
89279
89386
|
const iface = task.testInterface ?? [];
|
|
89280
89387
|
paths = [
|
|
@@ -89286,6 +89393,7 @@ function visibleAsk(task) {
|
|
|
89286
89393
|
...(task.kind !== "recovery" ? task.testTitles ?? [] : []).slice(0, TITLE_QUERIES).map(titleOfTest),
|
|
89287
89394
|
names.join(" ")
|
|
89288
89395
|
];
|
|
89396
|
+
hints = { files: paths, names: [...new Set(names)] };
|
|
89289
89397
|
}
|
|
89290
89398
|
queries = queries.map((q) => flat2(q).slice(0, QUERY_CHARS2).trim()).filter((q, i, all) => q !== "" && all.indexOf(q) === i).slice(0, MAX_QUERIES);
|
|
89291
89399
|
if (!queries.length && head) queries = [head];
|
|
@@ -89297,7 +89405,8 @@ function visibleAsk(task) {
|
|
|
89297
89405
|
task: [task.card.problem, task.card.acceptance].join("\n\n").slice(0, 1e3),
|
|
89298
89406
|
error: error2,
|
|
89299
89407
|
queries,
|
|
89300
|
-
...paths.length ? { paths } : {}
|
|
89408
|
+
...paths.length ? { paths } : {},
|
|
89409
|
+
hints
|
|
89301
89410
|
};
|
|
89302
89411
|
}
|
|
89303
89412
|
async function checkContext(deps, input) {
|
|
@@ -89339,24 +89448,30 @@ async function checkContext(deps, input) {
|
|
|
89339
89448
|
});
|
|
89340
89449
|
const state = stateOf(judged.knowledge);
|
|
89341
89450
|
if (!state || !judged.relevantIds) return null;
|
|
89451
|
+
const visible = new Set(reached.flatMap((c) => c.returnedIds));
|
|
89452
|
+
const count4 = (ids) => (ids ?? []).filter((id) => visible.has(id)).length;
|
|
89342
89453
|
return {
|
|
89343
89454
|
taskId: task.taskId,
|
|
89344
89455
|
inputsFingerprint: task.inputsFingerprint,
|
|
89345
89456
|
state,
|
|
89346
89457
|
relevantIds: judged.relevantIds,
|
|
89347
89458
|
judgeTag: judged.judgeTag,
|
|
89348
|
-
checkVersion: CONTEXT_CHECK_VERSION
|
|
89459
|
+
checkVersion: CONTEXT_CHECK_VERSION,
|
|
89460
|
+
onPoint: count4(judged.relevantIds),
|
|
89461
|
+
nearby: count4(judged.nearbyIds)
|
|
89349
89462
|
};
|
|
89350
89463
|
}
|
|
89351
89464
|
async function gateOnContext(deps, tasks, options) {
|
|
89352
89465
|
const states = /* @__PURE__ */ new Map();
|
|
89466
|
+
const scores = /* @__PURE__ */ new Map();
|
|
89353
89467
|
const checks = [];
|
|
89354
89468
|
for (const t of tasks)
|
|
89355
|
-
if (!needsCheck(t))
|
|
89356
|
-
|
|
89469
|
+
if (!needsCheck(t)) {
|
|
89470
|
+
states.set(t.taskId, t.context.state);
|
|
89471
|
+
scores.set(t.taskId, storedScoreOf(t));
|
|
89472
|
+
}
|
|
89357
89473
|
const toCheck = tasks.filter((t) => needsCheck(t));
|
|
89358
89474
|
await runPool(toCheck, options.concurrency, async (task) => {
|
|
89359
|
-
if (options.limit && found() >= options.limit) return;
|
|
89360
89475
|
const label = task.name ?? task.taskKey;
|
|
89361
89476
|
try {
|
|
89362
89477
|
await options.prepare?.(task);
|
|
@@ -89377,9 +89492,10 @@ async function gateOnContext(deps, tasks, options) {
|
|
|
89377
89492
|
return;
|
|
89378
89493
|
}
|
|
89379
89494
|
checks.push(check3);
|
|
89495
|
+
scores.set(task.taskId, { onPoint: check3.onPoint, nearby: check3.nearby });
|
|
89380
89496
|
const n = check3.relevantIds.length;
|
|
89381
89497
|
deps.log(
|
|
89382
|
-
`context ${check3.state.padEnd(6)} ${label}${n ? ` (${n}
|
|
89498
|
+
`context ${check3.state.padEnd(6)} ${label}${n ? ` (${n} on-point item${n === 1 ? "" : "s"}; returned: ${check3.onPoint} on point, ${check3.nearby} nearby)` : check3.nearby ? ` (${check3.nearby} nearby only)` : ""}`
|
|
89383
89499
|
);
|
|
89384
89500
|
} catch (err2) {
|
|
89385
89501
|
states.set(task.taskId, null);
|
|
@@ -89403,21 +89519,33 @@ async function gateOnContext(deps, tasks, options) {
|
|
|
89403
89519
|
`The context checks could not be reported (${err2 instanceof Error ? err2.message : String(err2)}); the next run checks them again.`
|
|
89404
89520
|
);
|
|
89405
89521
|
}
|
|
89406
|
-
const
|
|
89522
|
+
const found = [];
|
|
89407
89523
|
let noContext = 0;
|
|
89408
89524
|
let unknown2 = 0;
|
|
89409
89525
|
for (const t of tasks) {
|
|
89410
89526
|
const label = t.name ?? t.taskKey;
|
|
89411
|
-
if (!states.has(t.taskId)) continue;
|
|
89412
89527
|
const s = states.get(t.taskId);
|
|
89413
|
-
if (s === "found")
|
|
89528
|
+
if (s === "found") found.push(t);
|
|
89414
89529
|
else if (s === "missed" || s === "none") {
|
|
89415
89530
|
noContext += 1;
|
|
89416
|
-
deps.log(`skip ${label}: no
|
|
89531
|
+
deps.log(`skip ${label}: no on-point context Haansi can return (${s})`);
|
|
89417
89532
|
} else unknown2 += 1;
|
|
89418
89533
|
}
|
|
89534
|
+
const replay2 = rankFound(
|
|
89535
|
+
found,
|
|
89536
|
+
(t) => scores.get(t.taskId) ?? { onPoint: 0, nearby: 0 }
|
|
89537
|
+
);
|
|
89419
89538
|
return { replay: replay2, noContext, unknown: unknown2 };
|
|
89420
89539
|
}
|
|
89540
|
+
function fromChecks(tasks, limit, shard) {
|
|
89541
|
+
const ready = tasks.filter(
|
|
89542
|
+
(t) => !needsCheck(t) && t.context?.state === "found" && t.calibration?.state === "valid" && t.calibration.inputsFingerprint === t.inputsFingerprint
|
|
89543
|
+
);
|
|
89544
|
+
const ranked = rankFound(ready, storedScoreOf);
|
|
89545
|
+
const top = limit ? ranked.slice(0, limit) : ranked;
|
|
89546
|
+
const mine = shard && shard.count > 1 ? top.filter((_, i) => i % shard.count === shard.index - 1) : top;
|
|
89547
|
+
return { found: ready.length, top, mine };
|
|
89548
|
+
}
|
|
89421
89549
|
var CONTEXT_CHECK_VERSION, QUERY_CHARS2, ERROR_CHARS, MAX_PATHS, HEAD_CHARS, MAX_QUERIES, FAILING_QUERIES, TITLE_QUERIES, CONTEXT_REPORT_CHUNK, flat2;
|
|
89422
89550
|
var init_replay_context = __esm({
|
|
89423
89551
|
"src/commands/replay-context.ts"() {
|
|
@@ -89426,7 +89554,7 @@ var init_replay_context = __esm({
|
|
|
89426
89554
|
init_replay_diff();
|
|
89427
89555
|
init_replay_scale();
|
|
89428
89556
|
init_replay_spec();
|
|
89429
|
-
CONTEXT_CHECK_VERSION =
|
|
89557
|
+
CONTEXT_CHECK_VERSION = 4;
|
|
89430
89558
|
QUERY_CHARS2 = 200;
|
|
89431
89559
|
ERROR_CHARS = 1e3;
|
|
89432
89560
|
MAX_PATHS = 20;
|
|
@@ -89506,9 +89634,17 @@ function haansiSection(tools, hint = null) {
|
|
|
89506
89634
|
return [
|
|
89507
89635
|
"## Your organization's knowledge (Haansi)",
|
|
89508
89636
|
"",
|
|
89509
|
-
"These MCP servers hold what your organization learned before this task began
|
|
89637
|
+
"These MCP servers hold what your organization learned before this task began. They are read-only here.",
|
|
89638
|
+
"",
|
|
89639
|
+
"Start here, before you read or change any code, and search again when you get stuck:",
|
|
89510
89640
|
"",
|
|
89511
|
-
"
|
|
89641
|
+
"- Past fixes: search_solutions with a sentence about the problem (its error line in error, failing tests in failing_tests, the files you expect to change in files, identifiers in names), then get_solution to read the matching results in full.",
|
|
89642
|
+
"- Team procedures: match_playbook with the task, and search_skills; get_skill loads one.",
|
|
89643
|
+
"- Earlier attempts at the same code: get_prior_attempts with the files you expect to change.",
|
|
89644
|
+
"- Incidents like this one: search_prior_incidents.",
|
|
89645
|
+
"- Recent work, conventions and shared answers: search_digests, search_guides and search_exchange.",
|
|
89646
|
+
"",
|
|
89647
|
+
"A test name, an identifier or a list of file names alone rarely finds anything; ask with what is wrong, and pass those in the structured inputs. Read what applies and say which items you used.",
|
|
89512
89648
|
"",
|
|
89513
89649
|
...hint ? [HAANSI_HINTS[hint], ""] : [],
|
|
89514
89650
|
...tools.flatMap((s) => [
|
|
@@ -89667,13 +89803,26 @@ async function runStoredReplays(options, deps) {
|
|
|
89667
89803
|
);
|
|
89668
89804
|
return { results: [], notValid: 0, recorded: null };
|
|
89669
89805
|
}
|
|
89670
|
-
|
|
89671
|
-
|
|
89672
|
-
|
|
89673
|
-
|
|
89806
|
+
let mine;
|
|
89807
|
+
if (options.fromChecks) {
|
|
89808
|
+
const picked = fromChecks(set2.tasks, options.limit, options.shard ?? null);
|
|
89809
|
+
mine = picked.mine;
|
|
89674
89810
|
deps.log(
|
|
89675
|
-
`
|
|
89811
|
+
`From stored checks: ${picked.found} found task(s) in the set; the run takes ${picked.top.length}${options.limit ? ` (--limit ${options.limit})` : ""}, best scored first.`
|
|
89676
89812
|
);
|
|
89813
|
+
if (options.shard && options.shard.count > 1)
|
|
89814
|
+
deps.log(
|
|
89815
|
+
`Shard ${options.shard.index}/${options.shard.count}: ${mine.length} of the run's ${picked.top.length} task(s).`
|
|
89816
|
+
);
|
|
89817
|
+
if (mine.length === 0)
|
|
89818
|
+
return { results: [], notValid: 0, noContext: 0, recorded: null };
|
|
89819
|
+
} else {
|
|
89820
|
+
mine = set2.tasks.filter((t) => inShard(t.taskId, options.shard ?? null));
|
|
89821
|
+
if (options.shard && options.shard.count > 1)
|
|
89822
|
+
deps.log(
|
|
89823
|
+
`Shard ${options.shard.index}/${options.shard.count}: ${mine.length} of ${set2.tasks.length} task(s).`
|
|
89824
|
+
);
|
|
89825
|
+
}
|
|
89677
89826
|
const { states } = await runCalibration(
|
|
89678
89827
|
// With the context gate, --limit caps the tasks replayed (found ones),
|
|
89679
89828
|
// not the ones calibrated: calibrating only the first n left the rest
|
|
@@ -89753,7 +89902,6 @@ async function runStoredReplays(options, deps) {
|
|
|
89753
89902
|
agentCmd: options.judgeCmd || options.agentCmd,
|
|
89754
89903
|
timeoutMs: options.timeoutMs,
|
|
89755
89904
|
concurrency: options.concurrency ?? 1,
|
|
89756
|
-
limit: options.limit,
|
|
89757
89905
|
dryRun: options.dryRun,
|
|
89758
89906
|
cardOf: (task) => cardPrompt(task, null),
|
|
89759
89907
|
targetOf,
|
|
@@ -89764,6 +89912,19 @@ async function runStoredReplays(options, deps) {
|
|
|
89764
89912
|
noContext = gate.noContext;
|
|
89765
89913
|
contextUnknown = gate.unknown;
|
|
89766
89914
|
}
|
|
89915
|
+
if (options.checkOnly) {
|
|
89916
|
+
deps.log(
|
|
89917
|
+
`Checked: ${gated.length} found \xB7 ${noContext} with no on-point context \xB7 ${contextUnknown} unknown \xB7 ${notValid} not valid.`
|
|
89918
|
+
);
|
|
89919
|
+
return {
|
|
89920
|
+
results: [],
|
|
89921
|
+
notValid,
|
|
89922
|
+
noContext,
|
|
89923
|
+
contextUnknown,
|
|
89924
|
+
found: gated.length,
|
|
89925
|
+
recorded: null
|
|
89926
|
+
};
|
|
89927
|
+
}
|
|
89767
89928
|
const chosen = options.limit ? gated.slice(0, options.limit) : gated;
|
|
89768
89929
|
const results = [];
|
|
89769
89930
|
let recorded = 0;
|
|
@@ -90120,6 +90281,8 @@ function parseReplayArgs(argv) {
|
|
|
90120
90281
|
else if (takeScaleArg(a, next, scale, errors)) continue;
|
|
90121
90282
|
else if (a === "--force") options.force = true;
|
|
90122
90283
|
else if (a === "--any-context") options.anyContext = true;
|
|
90284
|
+
else if (a === "--check-only") options.checkOnly = true;
|
|
90285
|
+
else if (a === "--from-checks") options.fromChecks = true;
|
|
90123
90286
|
else if (a === "--max-hours") options.maxHours = Number(next());
|
|
90124
90287
|
else if (takeContainerArg(a, next, options.container, errors)) continue;
|
|
90125
90288
|
else if (a === "--arms")
|
|
@@ -90141,6 +90304,16 @@ function parseReplayArgs(argv) {
|
|
|
90141
90304
|
if (options.arms.length === 0 || options.arms.some((a) => !ARMS.includes(a)))
|
|
90142
90305
|
errors.push("--arms must list baseline and/or haansi.");
|
|
90143
90306
|
errors.push(...containerErrors(options.container));
|
|
90307
|
+
if (options.checkOnly && options.fromChecks)
|
|
90308
|
+
errors.push("--check-only and --from-checks are the two phases: pass one.");
|
|
90309
|
+
if ((options.checkOnly || options.fromChecks) && options.anyContext)
|
|
90310
|
+
errors.push(
|
|
90311
|
+
"--any-context skips the context check: not with --check-only or --from-checks."
|
|
90312
|
+
);
|
|
90313
|
+
if (options.checkOnly && options.limit !== null)
|
|
90314
|
+
errors.push(
|
|
90315
|
+
"--check-only checks every task: --limit belongs to the --from-checks phase."
|
|
90316
|
+
);
|
|
90144
90317
|
if (options.maxHours !== null && options.maxHours !== void 0 && (!Number.isFinite(options.maxHours) || options.maxHours <= 0))
|
|
90145
90318
|
errors.push("--max-hours must be a positive number.");
|
|
90146
90319
|
Object.assign(
|
|
@@ -90588,7 +90761,7 @@ async function replayStored(options, token2, apiUrl2) {
|
|
|
90588
90761
|
runCommand
|
|
90589
90762
|
);
|
|
90590
90763
|
stop = contained.stop;
|
|
90591
|
-
const { results, notValid, noContext, contextUnknown, recorded } = await runStoredReplays2(options, {
|
|
90764
|
+
const { results, notValid, noContext, contextUnknown, found, recorded } = await runStoredReplays2(options, {
|
|
90592
90765
|
...contained.deps,
|
|
90593
90766
|
getReplaySet: (repository) => apiGet3(
|
|
90594
90767
|
`/memory/replays/tasks?repository=${encodeURIComponent(repository)}`,
|
|
@@ -90636,6 +90809,13 @@ async function replayStored(options, token2, apiUrl2) {
|
|
|
90636
90809
|
apiUrl: apiUrl2,
|
|
90637
90810
|
token: token2
|
|
90638
90811
|
});
|
|
90812
|
+
if (options.checkOnly) {
|
|
90813
|
+
console.log(
|
|
90814
|
+
`
|
|
90815
|
+
Context checked: ${found ?? 0} task(s) found \xB7 ${noContext ?? 0} with no on-point context${contextUnknown ? ` \xB7 ${contextUnknown} could not be checked` : ""} \xB7 ${notValid} not valid to replay`
|
|
90816
|
+
);
|
|
90817
|
+
return;
|
|
90818
|
+
}
|
|
90639
90819
|
const arm = (a) => {
|
|
90640
90820
|
const mine = results.filter((r) => r.arm === a);
|
|
90641
90821
|
return `${mine.filter((r) => r.passed).length}/${mine.length}`;
|
|
@@ -90685,6 +90865,7 @@ var init_replay = __esm({
|
|
|
90685
90865
|
[--judge-cmd "<cmd>"]
|
|
90686
90866
|
[--run-ref <url>] [--dry-run] [--repo-commands] [--no-diffs]
|
|
90687
90867
|
[--shard i/n] [--concurrency k] [--force] [--max-hours h] [--any-context]
|
|
90868
|
+
[--check-only | --from-checks]
|
|
90688
90869
|
[--setup-cache <files> --setup-cache-path <paths> [--setup-cache-dir <dir>]]
|
|
90689
90870
|
[--container <image> [--container-setup "<cmd>"] [--container-env NAME]...]
|
|
90690
90871
|
|
|
@@ -90715,10 +90896,18 @@ files they are and import, and the files changed by the commits that wrote
|
|
|
90715
90896
|
the lines its stack frames name, or by those files' own last commits before
|
|
90716
90897
|
the start; a feature's problem, its tests' titles and the
|
|
90717
90898
|
names they use; the paths the card names), and the relevance judge (your agent
|
|
90718
|
-
command, once per task checked)
|
|
90719
|
-
|
|
90720
|
-
|
|
90721
|
-
|
|
90899
|
+
command, once per task checked) grades each item that came back on point (it
|
|
90900
|
+
says part of what to change or check), nearby (same area, another problem)
|
|
90901
|
+
or unrelated. A task is replayed only when an on-point item came back; one
|
|
90902
|
+
without is skipped, and checked again once the organization gains new Haansi
|
|
90903
|
+
content. Found tasks are replayed most on-point items first (then most
|
|
90904
|
+
nearby), and --limit takes the best. --any-context replays every valid task
|
|
90905
|
+
without the check (diagnostics).
|
|
90906
|
+
|
|
90907
|
+
Across shards, run two phases so --limit takes the best of the whole set:
|
|
90908
|
+
--check-only --shard i/n calibrates and checks each share (no agent runs);
|
|
90909
|
+
then --from-checks --limit N --shard i/n ranks every stored found task, takes
|
|
90910
|
+
the top N for the run, and replays this shard's share of them.
|
|
90722
90911
|
|
|
90723
90912
|
--judge-cmd runs the judges (whether the agent used what Haansi returned,
|
|
90724
90913
|
which of it was relevant, and the context check before replaying) with their
|
|
@@ -90737,7 +90926,7 @@ Run from a clone of the repository. Example (GitHub Actions):
|
|
|
90737
90926
|
haansi replay run --setup-cmd 'npm ci' \\
|
|
90738
90927
|
--container node:22 --container-env ANTHROPIC_API_KEY \\
|
|
90739
90928
|
--container-setup 'npm install -g haansi @anthropic-ai/claude-code' \\
|
|
90740
|
-
--agent-cmd 'claude -p "$HAANSI_REPLAY_PROMPT" --output-format stream-json --verbose --permission-mode acceptEdits --mcp-config "$HAANSI_REPLAY_MCP_CONFIG" --allowedTools mcp__haansi-solutions,mcp__haansi-exchange,mcp__haansi-digests,mcp__haansi-onboarding,mcp__haansi-skills' \\
|
|
90929
|
+
--agent-cmd 'claude -p "$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,mcp__haansi-incidents' \\
|
|
90741
90930
|
--test-cmd 'npx vitest run $HAANSI_REPLAY_TEST_FILES -t "$HAANSI_REPLAY_TEST_PATTERN"' \\
|
|
90742
90931
|
--agent-name claude-code --model claude-sonnet-5-5`;
|
|
90743
90932
|
}
|