${i} ${words}
`, + done: i % 5 === 0, + due_date: null, + priority: i % 4, + percent_done: 0, + bucket: i % 5 === 0 ? 12 : 11, + labels: i % 7 === 0 ? [9, 4] : [4], + assignees: [...(i % 3 === 0 ? [101] : []), ...(i % 4 === 0 ? [102] : []), ...(i % 2 === 0 ? [7] : [])], +}); +const GONE = new Set([13, 26]); + +function setup(t) { + const root = mkdtempSync(join(tmpdir(), 'bus-task-page-')); + const store = new Store(root); + const b = new Broker({ store, businesses }); + t.after(() => { + store.close(); + rmSync(root, { recursive: true, force: true }); + }); + const human = b.bindHuman({ business: 'demo', human: 'jason', via: 'cli', outsideAgent: true }); + const pm = b.bindLaunch({ business: 'demo', role: 'pm', run: 'demo-pm', harness: 'claude-code' }); + b.request(pm, { verb: 'role.claim' }); + b.recordTask({ + business: 'demo', + snapshots: Array.from({ length: N }, (_, k) => k + 1).map((i) => ({ + task_ref: ref(i), + updated: at, + digest, + fields: GONE.has(i) ? { gone: 'no-access' } : fieldsOf(i), + via: GONE.has(i) ? 'task' : 'board', + read_at: at, + })), + }); + // Task 2 and task 6 were created by the stack, under two requirements. + for (const [i, requirement] of [[2, 'REQ-SLICE-1'], [6, 'REQ-SLICE-2']]) { + const { request } = b.request(human, { verb: 'message.send', args: { to: 'pm', body: 'please' } }); + b.recordTask({ + cap: pm, + business: 'demo', + snapshots: [{ task_ref: ref(i), updated: at, digest, fields: fieldsOf(i) }], + events: [{ kind: 'task.created', subject: ref(i), body: { request, requirement } }], + }); + } + const tasks = (args, cap = pm) => b.request(cap, { verb: 'tasks', args }); + return { b, human, pm, reader: b.bindReader({ business: 'demo' }), tasks }; +} +const refusesWith = (code) => (e) => e.code === code; +// Every task, in the broker's order (task_ref, as text). +const ALL = Array.from({ length: N }, (_, k) => k + 1).sort((x, y) => (ref(x) < ref(y) ? -1 : 1)); +const live = (keep) => ALL.filter((i) => !GONE.has(i) && keep(i)).map(ref); +// Every page of a read, by following next. +function pages(tasks, args) { + const out = []; + let after; + for (;;) { + const page = tasks(after === undefined ? args : { ...args, after }); + out.push(page); + if (page.next === null) return out; + after = page.next; + } +} + +test('the session default fits under the reply cap and pages to the end with next', (t) => { + const { tasks } = setup(t); + const whole = tasks({}); + assert.equal(whole.length, N); + assert.ok(JSON.stringify(whole).length > 20 * REPLY_CAP, 'the whole board is far over the cap'); + const read = pages(tasks, { done: false }); + const open = live((i) => i % 5 !== 0); + assert.equal(read.length, Math.ceil(open.length / TASK_PAGE)); + for (const page of read) { + assert.ok(JSON.stringify(page).length < REPLY_CAP, `${JSON.stringify(page).length}`); + assert.equal(page.total, open.length); + assert.ok(page.tasks.length <= TASK_PAGE); + } + assert.deepEqual(read.flatMap((p) => p.tasks.map((x) => x.task_ref)), open); + assert.deepEqual(read.slice(0, -1).map((p) => p.next), read.slice(0, -1).map((p) => p.tasks.at(-1).task_ref)); + assert.equal(read.at(-1).next, null); + // limit at its most pages the same tasks in fewer pages. + const big = pages(tasks, { done: false, limit: TASK_PAGE_MAX }); + assert.equal(big.length, Math.ceil(open.length / TASK_PAGE_MAX)); + assert.deepEqual(big.flatMap((p) => p.tasks.map((x) => x.task_ref)), open); +}); + +test('the short form: no description, assignees by role, every listed key', (t) => { + const { tasks } = setup(t); + const whole = new Map(tasks({}).map((r) => [r.task_ref, r])); + const page = tasks({ limit: TASK_PAGE_MAX }); + for (const x of page.tasks) { + const row = whole.get(x.task_ref); + const i = Number(x.task_ref.split('/')[1]); + assert.deepEqual(x, { + task_ref: row.task_ref, + title: row.fields.title, + done: row.fields.done, + bucket: row.fields.bucket, + assignees: [...(i % 4 === 0 ? ['coder'] : []), ...(i % 3 === 0 ? ['pm'] : [])], + labels: row.fields.labels, + priority: row.priority, + requirement: row.requirement, + changed_at: row.at, + }); + } + const two = page.tasks.find((x) => x.task_ref === ref(2)); + assert.equal(two.requirement, 'REQ-SLICE-1'); + assert.equal(JSON.stringify(page).includes('description'), false); + // A tombstone is on the whole board, never in a filtered read. + assert.deepEqual(whole.get(ref(13)).fields, { gone: 'no-access' }); + const every = pages(tasks, { limit: TASK_PAGE_MAX }).flatMap((p) => p.tasks.map((x) => x.task_ref)); + assert.deepEqual(every, live(() => true)); +}); + +test('each filter, alone and together, and total counts every match', (t) => { + const { tasks } = setup(t); + const refs = (args) => pages(tasks, { ...args, limit: TASK_PAGE_MAX }).flatMap((p) => p.tasks.map((x) => x.task_ref)); + assert.deepEqual(refs({ done: true }), live((i) => i % 5 === 0)); + assert.deepEqual(refs({ done: false }), live((i) => i % 5 !== 0)); + assert.deepEqual(refs({ assignee: 'pm' }), live((i) => i % 3 === 0)); + assert.deepEqual(refs({ assignee: 'coder' }), live((i) => i % 4 === 0)); + // A role with no tracker bot holds no task; a person's assignment is no role's. + assert.deepEqual(refs({ assignee: 'reviewer' }), []); + assert.deepEqual(refs({ label: 9 }), live((i) => i % 7 === 0)); + assert.deepEqual(refs({ label: 4 }), live(() => true)); + assert.deepEqual(refs({ label: 5 }), []); + assert.deepEqual(refs({ requirement: 'REQ-SLICE-2' }), [ref(6)]); + assert.deepEqual(refs({ requirement: 'REQ-NONE-1' }), []); + assert.deepEqual(refs({ done: false, assignee: 'pm', label: 9 }), live((i) => i % 5 !== 0 && i % 3 === 0 && i % 7 === 0)); + const page = tasks({ done: true, assignee: 'coder', limit: 3 }); + assert.equal(page.total, live((i) => i % 5 === 0 && i % 4 === 0).length); + assert.equal(page.tasks.length, 3); + assert.equal(page.next, page.tasks[2].task_ref); + // after needn't be a task on the board, or one the filters match: the page starts past it, in + // text order (3/500 < 3/501 < 3/51), and total still counts every match. + const past = tasks({ after: 'vikunja:3/501', limit: 2 }); + assert.deepEqual(past.tasks.map((x) => x.task_ref), [ref(51), ref(52)]); + assert.equal(past.total, live(() => true).length); + assert.deepEqual(tasks({ done: true, after: ref(51), limit: 1 }).tasks.map((x) => x.task_ref), [ref(55)]); + const last = tasks({ after: 'vikunja:9/1' }); + assert.deepEqual(last, { tasks: [], total: live(() => true).length, next: null }); + // A page that ends exactly on the last match has no next. + const exact = tasks({ requirement: 'REQ-SLICE-1', limit: 1 }); + assert.deepEqual([exact.tasks.length, exact.next], [1, null]); +}); + +test('task_ref reads one task in full, as the whole board has it', (t) => { + const { tasks, human, reader } = setup(t); + const whole = tasks({}); + for (const r of [ref(2), ref(13), ref(500)]) assert.deepEqual(tasks({ task_ref: r }), whole.find((x) => x.task_ref === r)); + assert.ok(tasks({ task_ref: ref(2) }).fields.description.length > 2000); + assert.deepEqual(tasks({ task_ref: ref(2) }, human), tasks({ task_ref: ref(2) }, reader)); + assert.throws(() => tasks({ task_ref: 'vikunja:3/501' }), refusesWith('task-not-found')); + assert.throws(() => tasks({ task_ref: 'vikunja:4/2' }), refusesWith('task-not-found')); +}); + +test('no arguments is the read the Console and mosaic make, unchanged', (t) => { + const { b, tasks, human, reader } = setup(t); + const rows = b.taskView('demo', 'current'); + const whole = tasks({}); + assert.deepEqual(whole.map(({ priority, requirement, ...r }) => r), rows); + assert.deepEqual(tasks({}, human), whole); + assert.deepEqual(tasks({}, reader), whole); + assert.deepEqual(b.request(reader, { verb: 'tasks' }), whole); +}); + +test('an unknown argument or a bad value refuses invalid-request', (t) => { + const { tasks, reader } = setup(t); + const bad = [ + { status: 'open' }, + { description: 'x' }, + { done: 'false' }, + { done: 0 }, + { done: null }, + { assignee: 'cto' }, + { assignee: 'toString' }, + { assignee: 101 }, + { label: 0 }, + { label: -1 }, + { label: 1.5 }, + { label: '9' }, + { requirement: 'REQ-slice-1' }, + { requirement: 'REQ-SLICE-0' }, + { requirement: 7 }, + { limit: 0 }, + { limit: TASK_PAGE_MAX + 1 }, + { limit: 2.5 }, + { limit: '50' }, + { after: 'vikunja:3' }, + { after: 3 }, + { task_ref: 'vikunja:3/x' }, + { task_ref: ref(2), done: false }, + { task_ref: ref(2), limit: 1 }, + ]; + for (const args of bad) { + assert.throws(() => tasks(args), refusesWith('invalid-request'), JSON.stringify(args)); + assert.throws(() => tasks(args, reader), refusesWith('invalid-request'), JSON.stringify(args)); + } + assert.deepEqual([TASK_PAGE, TASK_PAGE_MAX], [50, 200]); +}); diff --git a/packages/harness/README.md b/packages/harness/README.md index 7598a573..1d1d7c5b 100644 --- a/packages/harness/README.md +++ b/packages/harness/README.md @@ -41,7 +41,10 @@ writes the bundle once into `dir` (0700; every file 0600, created with `wx`, so an existing bundle is never overwritten): - `prompt.md`: the role contract, then a generated "This session" section - (business, instance, run, who launched it, workspace, typed tools). + (business, instance, run, who launched it, workspace, built-in and typed + tools). The built-in tools are named as the harness names them, as the + manifest lists them: `limits.tools` for Pi, `claudeBuiltins` for Claude + Code (Read, Write, Edit, Grep, Glob). - `policy.json`: harness, workspace, built-in tools (`limits.tools`), typed tool names, network, the services the role needs, and the credential status the host read from the broker (metadata only, never a token). @@ -71,6 +74,20 @@ The harness never holds the capability. Tools call the runner's tool socket against `tools.json` and forwards the call with its own capability. A refusal comes back to the model as `refused:`.
+A reply goes to the model as compact JSON, from the MCP server and the Pi
+extension alike (`replyText` in `tools.mjs`). Over `REPLY_CAP`, 40000
+characters, the tool returns an error that names the reply's size and the
+cap and asks the session to narrow the query, never a cut-off reply. The
+pinned Claude CLI (2.1.296) saves an MCP result over 50000 characters to a
+file outside the workspace, which the gate refuses to read (row 89, #1583;
+`agents/dewey/work/queue-89/build.md` has how that size was found).
+
+`list_tasks` (tracker holders) takes the broker's `tasks` filters, `done`,
+`assignee`, `label`, `requirement`, `limit` and `after`, and sends
+`done: false` when the session leaves `done` out, so its default is open
+tasks, 50 at a time, without descriptions. `get_task` takes a `task_ref`
+and returns that task in full. `packages/bus/README.md` has the reply.
+
## The gate
`gate.mjs` `decide(policy, tool, input)` answers one tool call: the
@@ -188,6 +205,13 @@ These are limits, not bugs, and nothing in this package claims otherwise.
- **Resolution** runs without the project layer, there is no skills source
(bundles list none), and the resolved `thinking` level is recorded in the
manifest but not applied to either harness.
+- **The reply cap is per reply.** Claude Code also persists the largest
+ results of one message when their total passes 200000 characters, so
+ six or more replies near the cap in one turn can still be saved to a
+ file. A single task whose full row is over the cap (a description near
+ 40000 characters) can't be read through `get_task`. The cap matches
+ Claude CLI 2.1.296; the binary the bus unit names updates itself, so a
+ later version can move the threshold.
- **`--restricted`** (Claude Code) is not one of the S0 lines. It is what
keeps `CLAUDE.md` files and auto-memory out of the prompt (above).
- **Other spellings.** A read is refused when a spelling Pi might open
diff --git a/packages/harness/src/bundle.mjs b/packages/harness/src/bundle.mjs
index 2f16b1c8..9326be87 100644
--- a/packages/harness/src/bundle.mjs
+++ b/packages/harness/src/bundle.mjs
@@ -64,7 +64,8 @@ function prompt({ contract, resolved, session, tools, untyped }) {
`- Business: ${resolved.business}. Role instance: ${resolved.instance} (definition ${resolved.definition}). Run: ${session.run}.`,
`- Launched by: ${session.launchedBy}. Harness: ${session.harness}, model ${session.model}.`,
`- Workspace: ${session.workspace}. File tools work inside it only.`,
- `- Built-in tools: ${resolved.limits.tools.length ? resolved.limits.tools.join(", ") : "none"}.`,
+ // Row 89 (#1583): the names the harness gives its built-ins, as the manifest lists them.
+ `- Built-in tools: ${(session.harness === "pi" ? resolved.limits.tools : claudeBuiltins({ tools: resolved.limits.tools })).join(", ") || "none"}.`,
`- Bus tools: ${tools.map((t) => t.name).join(", ")}. The broker decides every call; a refusal names its reason.`,
"- Each message to you arrives as one turn. Your final answer goes back to the sender as the reply, so end each turn with the answer itself.",
"- You never hold a token. You act on the tracker, the forge and other roles only through the bus tools.",
diff --git a/packages/harness/src/mcp-server.mjs b/packages/harness/src/mcp-server.mjs
index 0f43948e..5415a55e 100644
--- a/packages/harness/src/mcp-server.mjs
+++ b/packages/harness/src/mcp-server.mjs
@@ -1,11 +1,12 @@
// Minimal stdio MCP server that gives a Claude Code session its typed
// tools: `node mcp-server.mjs `. Newline-delimited
// JSON-RPC 2.0 with initialize, tools/list, tools/call and ping. Each call
-// goes to the runner's tool socket; the server holds no capability.
+// goes to the runner's tool socket; the server holds no capability. A
+// reply over REPLY_CAP is a tool error (tools.mjs, replyText).
import { readFileSync } from "node:fs";
import { createInterface } from "node:readline";
-import { callTool } from "./tools.mjs";
+import { callTool, replyText } from "./tools.mjs";
const [toolsFile, socket] = process.argv.slice(2);
if (!toolsFile || !socket) {
@@ -37,7 +38,7 @@ async function handle(m) {
if (!byName.has(name)) return error(id, -32602, `unknown tool: ${name}`);
try {
const result = await callTool(socket, name, m.params?.arguments ?? {});
- return reply(id, { content: [{ type: "text", text: JSON.stringify(result, null, 2) }] });
+ return reply(id, { content: [{ type: "text", text: replyText(result) }] });
} catch (e) {
return reply(id, { content: [{ type: "text", text: e.message }], isError: true });
}
diff --git a/packages/harness/src/pi-extension.mjs b/packages/harness/src/pi-extension.mjs
index 2b3b0044..5647f0f5 100644
--- a/packages/harness/src/pi-extension.mjs
+++ b/packages/harness/src/pi-extension.mjs
@@ -10,7 +10,7 @@
import { readFileSync, writeFileSync } from "node:fs";
import { decide } from "./gate.mjs";
-import { callTool } from "./tools.mjs";
+import { callTool, replyText } from "./tools.mjs";
function need(name) {
const value = process.env[name];
@@ -40,9 +40,9 @@ export default function (pi) {
description: t.description,
parameters: t.parameters,
async execute(_toolCallId, params) {
- // A refusal throws, which pi reports to the model as a tool error.
+ // A refusal, or a reply over REPLY_CAP, throws, which pi reports to the model as a tool error.
const result = await callTool(socket, t.name, params ?? {});
- return { content: [{ type: "text", text: JSON.stringify(result, null, 2) }], details: {} };
+ return { content: [{ type: "text", text: replyText(result) }], details: {} };
},
});
}
diff --git a/packages/harness/src/tools.mjs b/packages/harness/src/tools.mjs
index 903f3bef..f0011d30 100644
--- a/packages/harness/src/tools.mjs
+++ b/packages/harness/src/tools.mjs
@@ -16,6 +16,11 @@ const str = (description, extra = {}) => ({ type: "string", description, ...extr
const obj = (properties, required = []) => ({ type: "object", properties, required, additionalProperties: false });
const TASK_REF = str("Task reference, as list_tasks shows it");
+// The broker's page size for `tasks`, by default and at most (TASK_PAGE and TASK_PAGE_MAX in
+// packages/bus/src/broker.mjs; a test holds them equal). Kept here so the session's tool server
+// doesn't load the broker.
+const TASK_PAGE = 50;
+const TASK_PAGE_MAX = 200;
const COMMON = {
task_ref: TASK_REF,
decision: str("Id of a resolved decision that approves this call; needed when the action is cross-role or gated for you"),
@@ -113,10 +118,39 @@ export function typedTools({ resolved, arbiter = false, tracker = false }) {
add("list_agents", "agents", null, "List the role instances that hold a claim now.", obj({}));
add("inbox", "inbox", null, "List the open decisions routed to you.", obj({}));
add("trail", "trail", null, "Show the evidence trail for a subject: a run, message, decision or task.", obj({ subject: str("Subject id") }, ["subject"]));
- if (tracker) add("list_tasks", "tasks", null, "List the business's tracker tasks as the broker last saw them.", obj({}));
+ if (tracker) {
+ add("list_tasks", "tasks", null, `List the business's tracker tasks as the broker last saw them, in short form with no description. By default open tasks (done false), ${TASK_PAGE} at a time; pass the reply's next as after for the next page. get_task reads one task in full.`, obj({
+ done: { type: "boolean", description: "true for closed tasks, false (the default) for open ones" },
+ assignee: str("Only tasks assigned to this role"),
+ label: { type: "integer", minimum: 1, description: "Only tasks with this label id" },
+ requirement: str("Only tasks created for this requirement id", { pattern: "^REQ-[A-Z]+-[1-9][0-9]*$" }),
+ limit: { type: "integer", minimum: 1, maximum: TASK_PAGE_MAX, description: `Tasks per page, ${TASK_PAGE} by default` },
+ after: str("The next value of the previous page"),
+ }));
+ add("get_task", "tasks", null, "Read one tracker task in full, description included.", obj({ task_ref: TASK_REF }, ["task_ref"]));
+ }
return tools;
}
+// Arguments a tool sends when the session leaves them out. list_tasks always filters: the
+// broker's `tasks` with no arguments is the whole board, which is the Console's read.
+const ARG_DEFAULTS = { list_tasks: { done: false } };
+
+// The most characters a tool reply may hold. The pinned Claude CLI (2.1.296, decision 91) saves
+// an MCP result over 50000 characters to a file outside the workspace, which the run's file guard
+// refuses to read (#1583; agents/dewey/work/queue-89/build.md has how that was found). The cap keeps
+// 20% under it, counts as JavaScript does (UTF-16 units, as the CLI does), and holds for pi too.
+export const REPLY_CAP = 40_000;
+
+// The text a tool returns to the model: compact JSON, or an error over the cap. A reply is never
+// cut short.
+export function replyText(result) {
+ const text = JSON.stringify(result) ?? "null";
+ if (text.length > REPLY_CAP)
+ throw new Error(`reply too large: ${text.length} characters, over the ${REPLY_CAP}-character cap. Narrow the query: a filter, a smaller limit, or get_task for one task.`);
+ return text;
+}
+
// Held actions that have no typed tool. The manifest lists them so a
// reviewer sees what the session can't do through the bus.
export function untypedActions(resolved, tools) {
@@ -137,7 +171,7 @@ export function callTool(path, tool, args = {}, { timeout = 60_000 } = {}) {
error ? reject(error) : resolve(value);
};
socket.setTimeout(timeout, () => end(new Error("tool call timed out")));
- socket.on("connect", () => socket.write(JSON.stringify({ tool, args }) + "\n"));
+ socket.on("connect", () => socket.write(JSON.stringify({ tool, args: { ...ARG_DEFAULTS[tool], ...args } }) + "\n"));
socket.on("error", (e) => end(new Error(`tool socket: ${e.code ?? e.message}`)));
socket.on("end", () => end(new Error("tool socket closed without a reply")));
socket.on("data", (b) => {
diff --git a/packages/harness/tests/bundle.test.mjs b/packages/harness/tests/bundle.test.mjs
index 23deabf2..98cee8ed 100644
--- a/packages/harness/tests/bundle.test.mjs
+++ b/packages/harness/tests/bundle.test.mjs
@@ -87,6 +87,20 @@ test("a claude-code bundle adds the wrapped gate hook and the MCP config", (t) =
assert.deepEqual(Object.keys(m.sources).sort(), ["claudeGate", "gate", "mcpServer", "tools"]);
});
+// Row 89 (#1583): the prompt names the built-ins as the session's harness names them.
+test("the prompt's built-in tools line is the harness's own names, as the manifest lists them", (t) => {
+ const line = (b) => readFileSync(b.files.prompt, "utf8").split("\n").find((l) => l.startsWith("- Built-in tools: "));
+ const pi = build(t, resolvedFor(t, "pm"));
+ assert.equal(line(pi), "- Built-in tools: read, write, edit, grep, find, ls.");
+ const claude = build(t, resolvedFor(t, "pm", (d) => (d.roles.pm.vars = { ...d.roles.pm.vars, harness: "claude-code", model: "claude-opus-5-5" })));
+ assert.equal(line(claude), "- Built-in tools: Read, Write, Edit, Grep, Glob.");
+ for (const b of [pi, claude]) assert.equal(line(b), `- Built-in tools: ${b.manifest.builtins.join(", ")}.`);
+ for (const harness of ["pi", "claude-code"]) {
+ const none = build(t, resolvedFor(t, "pm", (d) => (d.roles.pm.vars = { ...d.roles.pm.vars, harness, model: "claude-opus-5-5", "limits.tools": [] })));
+ assert.equal(line(none), "- Built-in tools: none.", harness);
+ }
+});
+
test("a bundle is written once: an existing file refuses", (t) => {
const dir = scratch(t);
mkdirSync(join(dir, "bundle"));
diff --git a/packages/harness/tests/mcp-server.test.mjs b/packages/harness/tests/mcp-server.test.mjs
index 72992dde..2219730c 100644
--- a/packages/harness/tests/mcp-server.test.mjs
+++ b/packages/harness/tests/mcp-server.test.mjs
@@ -5,7 +5,7 @@ import { writeFileSync } from "node:fs";
import { join } from "node:path";
import { createInterface } from "node:readline";
import { SOURCES } from "../src/bundle.mjs";
-import { typedTools } from "../src/tools.mjs";
+import { REPLY_CAP, typedTools } from "../src/tools.mjs";
import { fakeToolSocket, resolvedFor, scratch } from "./helpers.mjs";
// The server over stdio: send(message) resolves with the reply to its id.
@@ -43,6 +43,8 @@ async function setup(t) {
writeFileSync(toolsFile, JSON.stringify(tools));
const sock = await fakeToolSocket(t, dir, (tool, args) => {
if (tool === "launch") throw new Error("capacity-full");
+ // trail answers a string `size` characters long once JSON-encoded, for the reply cap.
+ if (tool === "trail" && /^size:/.test(args.subject)) return "x".repeat(Number(args.subject.slice(5)) - 2);
return { tool, args };
});
return { dir, tools, sock, server: serve(t, toolsFile, sock.path) };
@@ -70,6 +72,25 @@ test("tools/call goes through the tool socket; a refusal is an isError result",
assert.deepEqual(sock.calls.map((c) => c.tool), ["list_agents", "launch"]);
});
+test("replies are compact JSON; over the cap a tool error names the size and the cap", async (t) => {
+ const { sock, server } = await setup(t);
+ const call = async (name, args) => (await server.send("tools/call", { name, arguments: args })).result;
+ const ok = await call("get_task", { task_ref: "vikunja:3/7" });
+ assert.equal(ok.content[0].text, '{"tool":"get_task","args":{"task_ref":"vikunja:3/7"}}');
+ // list_tasks with no arguments asks the broker for open tasks.
+ assert.equal((await call("list_tasks", {})).content[0].text, '{"tool":"list_tasks","args":{"done":false}}');
+ const fits = await call("trail", { subject: `size:${REPLY_CAP}` });
+ assert.equal(fits.isError, undefined);
+ assert.equal(fits.content[0].text.length, REPLY_CAP);
+ const over = await call("trail", { subject: `size:${REPLY_CAP + 1}` });
+ assert.equal(over.isError, true);
+ assert.deepEqual(over.content, [{
+ type: "text",
+ text: `reply too large: ${REPLY_CAP + 1} characters, over the ${REPLY_CAP}-character cap. Narrow the query: a filter, a smaller limit, or get_task for one task.`,
+ }]);
+ assert.deepEqual(sock.calls.map((c) => c.tool), ["get_task", "list_tasks", "trail", "trail"]);
+});
+
test("unknown tools and methods are JSON-RPC errors and never reach the socket", async (t) => {
const { sock, server } = await setup(t);
const unknown = await server.send("tools/call", { name: "resolve_decision", arguments: {} });
diff --git a/packages/harness/tests/pi-session.test.mjs b/packages/harness/tests/pi-session.test.mjs
index 32db2aa3..95954b9d 100644
--- a/packages/harness/tests/pi-session.test.mjs
+++ b/packages/harness/tests/pi-session.test.mjs
@@ -8,6 +8,7 @@ import { existsSync, mkdirSync, readdirSync, symlinkSync, writeFileSync } from "
import { join } from "node:path";
import { buildBundle } from "../src/bundle.mjs";
import { adapterEnv } from "../src/runner.mjs";
+import { REPLY_CAP } from "../src/tools.mjs";
import { fakeToolSocket, mockAnthropic, REPO, resolvedFor, scratch } from "./helpers.mjs";
const ADAPTER = join(REPO, "adapters", "pi", "adapter.sh");
@@ -25,6 +26,8 @@ async function session(t, script, place = (dir) => join(dir, "ws")) {
const resolved = resolvedFor(t, "pm", (d) => (d.roles.pm.vars = { ...d.roles.pm.vars, model: "probe-model" }));
const sock = await fakeToolSocket(t, dir, (tool, args) => {
if (tool === "launch") throw new Error("capacity-full");
+ // trail answers a string `size` characters long once JSON-encoded, for the reply cap.
+ if (tool === "trail" && /^size:/.test(args.subject)) return "x".repeat(Number(args.subject.slice(5)) - 2);
return { tool, args };
});
const s = {
@@ -78,7 +81,7 @@ test("pi: typed tools reach the socket, the gate blocks, agent_end writes the ma
assert.equal(results.length, 5);
// list_agents went through the socket; launch was refused by it.
assert.equal(results[0][0], false);
- assert.match(results[0][1], /"tool": "list_agents"/);
+ assert.match(results[0][1], /^\{"tool":"list_agents"/);
assert.equal(results[1][0], true);
assert.match(results[1][1], /refused: capacity-full/);
// read inside passes; outside is gated; bash isn't in the PM's tools.
@@ -96,6 +99,30 @@ test("pi: typed tools reach the socket, the gate blocks, agent_end writes the ma
assert.ok(readdirSync(s.sessionDir).length > 0);
});
+test("pi: a reply over the cap is a tool error naming the size and the cap; list_tasks asks for open tasks", async (t) => {
+ const { sock, s, env } = await session(t, [
+ { name: "trail", input: { subject: `size:${REPLY_CAP}` } },
+ { name: "trail", input: { subject: `size:${REPLY_CAP + 1}` } },
+ { name: "list_tasks", input: {} },
+ { name: "get_task", input: { task_ref: "vikunja:3/7" } },
+ ]);
+ const r = await turn(env, "Message 1 from jason, class REQUEST:\n\nshow the board", s.workspace);
+ assert.equal(r.code, 0, r.stderr);
+ const results = JSON.parse(r.stdout.trim().slice("ANSWER ".length));
+ assert.equal(results.length, 4);
+ assert.deepEqual(results[0], [false, `"${"x".repeat(79)}`]);
+ assert.equal(results[1][0], true);
+ assert.match(results[1][1], new RegExp(`^reply too large: ${REPLY_CAP + 1} characters, over the ${REPLY_CAP}-character cap`));
+ assert.deepEqual(results[2], [false, '{"tool":"list_tasks","args":{"done":false}}']);
+ assert.deepEqual(results[3], [false, '{"tool":"get_task","args":{"task_ref":"vikunja:3/7"}}']);
+ assert.deepEqual(sock.calls.map((c) => [c.tool, c.args]), [
+ ["trail", { subject: `size:${REPLY_CAP}` }],
+ ["trail", { subject: `size:${REPLY_CAP + 1}` }],
+ ["list_tasks", { done: false }],
+ ["get_task", { task_ref: "vikunja:3/7" }],
+ ]);
+});
+
test("pi: a write through a dangling symlink is blocked, and nothing appears outside", async (t) => {
const { dir, workspace, s, env } = await session(t, [
{ name: "write", input: { path: "notes.md", content: "planted\n" } },
diff --git a/packages/harness/tests/tools.test.mjs b/packages/harness/tests/tools.test.mjs
index 3fe7dd1e..045f11c1 100644
--- a/packages/harness/tests/tools.test.mjs
+++ b/packages/harness/tests/tools.test.mjs
@@ -1,6 +1,7 @@
import { test } from "node:test";
import assert from "node:assert/strict";
-import { callTool, typedTools, untypedActions } from "../src/tools.mjs";
+import { callTool, REPLY_CAP, replyText, typedTools, untypedActions } from "../src/tools.mjs";
+import { TASK_PAGE, TASK_PAGE_MAX } from "../../bus/src/broker.mjs";
import { fakeToolSocket, resolvedFor, scratch } from "./helpers.mjs";
const names = (tools) => tools.map((t) => t.name);
@@ -12,7 +13,7 @@ test("the PM gets launch, its task verbs and the reads", (t) => {
"send_message", "launch",
"task_create", "task_assign", "task_reassign", "task_schedule", "task_priority_change", "task_scope_change",
"task_update_assigned", "task_close",
- "raise_decision", "resolve_decision", "list_agents", "inbox", "trail", "list_tasks",
+ "raise_decision", "resolve_decision", "list_agents", "inbox", "trail", "list_tasks", "get_task",
]);
const launch = tools.find((x) => x.name === "launch");
assert.equal(launch.verb, "launch");
@@ -37,7 +38,7 @@ test("a coder with a tracker gets its own task tools and no task_reassign", (t)
const coder = resolvedFor(t, "coder");
assert.deepEqual(names(typedTools({ resolved: coder, tracker: true })), [
"send_message", "task_scope_change", "task_update_assigned",
- "raise_decision", "list_agents", "inbox", "trail", "list_tasks",
+ "raise_decision", "list_agents", "inbox", "trail", "list_tasks", "get_task",
]);
});
@@ -49,7 +50,7 @@ test("launch only when the business's launch block names the instance as launche
test("an action outside the instance's authority has no tool", (t) => {
const coder = resolvedFor(t, "coder", (d) => (d.roles.coder.vars["limits.authority"] = ["review.request"]));
- assert.deepEqual(names(typedTools({ resolved: coder, tracker: true })), ["raise_decision", "list_agents", "inbox", "trail", "list_tasks"]);
+ assert.deepEqual(names(typedTools({ resolved: coder, tracker: true })), ["raise_decision", "list_agents", "inbox", "trail", "list_tasks", "get_task"]);
});
test("callTool: one JSON line out, the result back, a refusal rejects", async (t) => {
@@ -63,3 +64,54 @@ test("callTool: one JSON line out, the result back, a refusal rejects", async (t
await assert.rejects(callTool(`${dir}/missing.sock`, "x"), /tool socket: ENOENT/);
assert.deepEqual(sock.calls.map((c) => c.tool), ["list_agents", "launch"]);
});
+
+// Row 89 (#1583): task reads a session can page, and replies that fit.
+test("list_tasks takes the broker's filters and pages; get_task takes a task_ref", (t) => {
+ const tools = typedTools({ resolved: resolvedFor(t, "pm"), tracker: true });
+ const list = tools.find((x) => x.name === "list_tasks");
+ const get = tools.find((x) => x.name === "get_task");
+ assert.deepEqual([list.verb, list.action, get.verb, get.action], ["tasks", null, "tasks", null]);
+ const p = list.parameters;
+ assert.deepEqual(Object.keys(p.properties), ["done", "assignee", "label", "requirement", "limit", "after"]);
+ assert.deepEqual(p.required, []);
+ assert.equal(p.properties.done.type, "boolean");
+ assert.equal(p.properties.assignee.type, "string");
+ assert.deepEqual([p.properties.label.type, p.properties.label.minimum], ["integer", 1]);
+ assert.equal(p.properties.requirement.pattern, "^REQ-[A-Z]+-[1-9][0-9]*$");
+ assert.equal(p.properties.after.type, "string");
+ // The page sizes are the broker's own.
+ assert.deepEqual([p.properties.limit.type, p.properties.limit.minimum, p.properties.limit.maximum], ["integer", 1, TASK_PAGE_MAX]);
+ assert.match(p.properties.limit.description, new RegExp(`\\b${TASK_PAGE} by default`));
+ assert.match(list.description, new RegExp(`open tasks \\(done false\\), ${TASK_PAGE} at a time`));
+ assert.match(list.description, /next as after/);
+ assert.match(list.description, /get_task/);
+ assert.deepEqual(get.parameters.properties, { task_ref: tools.find((x) => x.name === "task_close").parameters.properties.task_ref });
+ assert.deepEqual(get.parameters.required, ["task_ref"]);
+ // Neither without a tracker.
+ assert.deepEqual(names(typedTools({ resolved: resolvedFor(t, "pm") })).filter((n) => n.endsWith("_tasks") || n === "get_task"), []);
+});
+
+test("callTool sends list_tasks done false unless the session names done; other tools as given", async (t) => {
+ const sock = await fakeToolSocket(t, scratch(t), (tool, args) => args);
+ assert.deepEqual(await callTool(sock.path, "list_tasks"), { done: false });
+ assert.deepEqual(await callTool(sock.path, "list_tasks", { limit: 5 }), { done: false, limit: 5 });
+ assert.deepEqual(await callTool(sock.path, "list_tasks", { done: true }), { done: true });
+ assert.deepEqual(await callTool(sock.path, "get_task", { task_ref: "vikunja:3/7" }), { task_ref: "vikunja:3/7" });
+ assert.deepEqual(await callTool(sock.path, "list_agents"), {});
+});
+
+test("replyText: compact JSON up to the cap, an error naming the size and the cap over it", () => {
+ assert.equal(REPLY_CAP, 40_000);
+ assert.equal(replyText({ a: [1, { b: "c" }] }), '{"a":[1,{"b":"c"}]}');
+ assert.equal(replyText(null), "null");
+ assert.equal(replyText(undefined), "null");
+ // A JSON string adds its two quotes; length counts UTF-16 units, as the CLI does.
+ assert.equal(replyText("x".repeat(REPLY_CAP - 2)).length, REPLY_CAP);
+ assert.throws(() => replyText("x".repeat(REPLY_CAP - 1)), (e) => {
+ assert.equal(e.message, `reply too large: ${REPLY_CAP + 1} characters, over the ${REPLY_CAP}-character cap. Narrow the query: a filter, a smaller limit, or get_task for one task.`);
+ return true;
+ });
+ assert.throws(() => replyText("\u{1F600}".repeat(REPLY_CAP / 2)), /reply too large: 40002 characters/);
+ // 40000 units fit, though they are 80002 bytes: the cap isn't a byte count.
+ assert.equal(replyText("\u{1F600}".repeat(REPLY_CAP / 2 - 1)).length, REPLY_CAP);
+});