ZacharyZhang-NY/AgentCompany
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 2 Wiki Activity

Q40: tasks choose a writing mode; house style applies to conversational work only

Browse Source 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
ZacharyZhang-NYandClaude Opus 5.5 committed 2026-10-08 07:28:47 -04:00
1 parent 645ce686bb
commit 7be072f8c5
18 files changed
+119 -41
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

tests/integration/collaboration-flow.test.ts
+1 -1
View File
@@ -98,7 +98,7 @@ describe("collaboration integration flow", () => {
revisionId = skills.importDirectory({ orgId, directoryPath: createSkillDirectory(), versionSemver: "1.0.0" });
skillId = (connection.db.prepare("select skill_id from skill_revisions where id = ?").get(revisionId) as { skill_id: string }).skill_id;
skills.bind({ orgId, agentId, skillId, revisionId });
skills.resolveRuntimeSnapshot({ orgId, agentId, runId: `run_${randomUUID().replaceAll("-", "")}`, maxTokens: 44_800, taskText: "", runModel: { providerKey: "openai", model: "gpt-5.4" } });
skills.resolveRuntimeSnapshot({ orgId, agentId, runId: `run_${randomUUID().replaceAll("-", "")}`, maxTokens: 44_800, taskText: "", runModel: { providerKey: "openai", model: "gpt-5.4" }, houseStyle: true });
suiteId = (connection.db.prepare("select id from eval_suites where org_id = ? and key = ?")
.get(orgId, EVAL_SUITE_KEY.INTEGRATION) as { id: string }).id;
evalRunId = sdlc.runEvalSuite({ orgId, suiteId });
tests/unit/builtin-skills.test.ts
+14 -7
View File
@@ -171,7 +171,7 @@ describe("built-in skills", () => {
h.skills.bind({ orgId: h.orgId, agentId, skillId: skill.id, revisionId: skill.revision_id });
});
const prompt = (runId: string) => {
const id = h.connection.transaction(() => h.skills.resolveRuntimeSnapshot({ orgId: h.orgId, agentId, runId, maxTokens: SKILL_TOKENS, taskText: "", runModel: { providerKey: "openai", model: "gpt-5.4" } }));
const id = h.connection.transaction(() => h.skills.resolveRuntimeSnapshot({ orgId: h.orgId, agentId, runId, maxTokens: SKILL_TOKENS, taskText: "", runModel: { providerKey: "openai", model: "gpt-5.4" }, houseStyle: true }));
return (h.connection.db.prepare("select compiled_prompt from skill_runtime_snapshots where id = ?").get(id) as { compiled_prompt: string }).compiled_prompt;
};
expect(prompt("run_before")).toContain("# Skill: Backend Architect");
@@ -265,7 +265,7 @@ describe("built-in skills", () => {
expect(hired).toContain("Unknown skills skipped: no-such-skill");
const danaId = (h.connection.db.prepare("select id from agents where org_id = ? and name = 'Dana Lee'").get(h.orgId) as { id: string }).id;
const danaSnapshot = h.connection.transaction(() => h.skills.resolveRuntimeSnapshot({ orgId: h.orgId, agentId: danaId, runId: "run_dana", maxTokens: SKILL_TOKENS, taskText: "", runModel: { providerKey: "openai", model: "gpt-5.4" } }));
const danaSnapshot = h.connection.transaction(() => h.skills.resolveRuntimeSnapshot({ orgId: h.orgId, agentId: danaId, runId: "run_dana", maxTokens: SKILL_TOKENS, taskText: "", runModel: { providerKey: "openai", model: "gpt-5.4" }, houseStyle: true }));
const danaPrompt = (h.connection.db.prepare("select compiled_prompt from skill_runtime_snapshots where id = ?").get(danaSnapshot) as { compiled_prompt: string }).compiled_prompt;
expect(danaPrompt.indexOf("# Skill: humanizer")).toBe(0);
expect(danaPrompt).toContain("# Skill: Backend Architect");
@@ -274,12 +274,19 @@ describe("built-in skills", () => {
expect(danaPrompt.length).toBeLessThan(SKILL_TOKENS * 3 + 5_000);
// The CEO carries humanizer too - forced by resolution, not by binding.
expect(h.skills.hasRuntimeSkills(h.orgId, ceoId)).toBe(true);
const ceoSnapshot = h.connection.transaction(() => h.skills.resolveRuntimeSnapshot({ orgId: h.orgId, agentId: ceoId, runId: "run_ceo", maxTokens: SKILL_TOKENS, taskText: "", runModel: { providerKey: "openai", model: "gpt-5.4" } }));
expect(h.skills.hasRuntimeSkills(h.orgId, ceoId, true)).toBe(true);
const ceoSnapshot = h.connection.transaction(() => h.skills.resolveRuntimeSnapshot({ orgId: h.orgId, agentId: ceoId, runId: "run_ceo", maxTokens: SKILL_TOKENS, taskText: "", runModel: { providerKey: "openai", model: "gpt-5.4" }, houseStyle: true }));
const ceoPrompt = (h.connection.db.prepare("select compiled_prompt from skill_runtime_snapshots where id = ?").get(ceoSnapshot) as { compiled_prompt: string }).compiled_prompt;
expect(ceoPrompt).toContain("# Skill: humanizer");
expect(ceoPrompt).not.toContain("# Skill: Backend Architect");
// Technical and structured work drops house style; a CEO without grants then carries nothing.
expect(h.skills.hasRuntimeSkills(h.orgId, ceoId, false)).toBe(false);
const technicalSnapshot = h.connection.transaction(() => h.skills.resolveRuntimeSnapshot({ orgId: h.orgId, agentId: danaId, runId: "run_dana_spec", maxTokens: SKILL_TOKENS, taskText: "", runModel: { providerKey: "openai", model: "gpt-5.4" }, houseStyle: false }));
const technicalPrompt = (h.connection.db.prepare("select compiled_prompt from skill_runtime_snapshots where id = ?").get(technicalSnapshot) as { compiled_prompt: string }).compiled_prompt;
expect(technicalPrompt).not.toContain("# Skill: humanizer");
expect(technicalPrompt.indexOf("# Skill: Backend Architect")).toBe(0);
// list_skills shows descriptions and marks the house-style skill.
const listing = dispatchAgentTool(deps, ceoCtx, { id: "call_2", name: "list_skills", arguments: {} });
expect(listing).toContain("humanizer (humanizer)");
@@ -300,7 +307,7 @@ describe("built-in skills", () => {
});
});
const promptFor = (runId: string) => {
const snapshotId = h.connection.transaction(() => h.skills.resolveRuntimeSnapshot({ orgId: h.orgId, agentId, runId, maxTokens: SKILL_TOKENS, taskText: "", runModel: { providerKey: "openai", model: "gpt-5.4" } }));
const snapshotId = h.connection.transaction(() => h.skills.resolveRuntimeSnapshot({ orgId: h.orgId, agentId, runId, maxTokens: SKILL_TOKENS, taskText: "", runModel: { providerKey: "openai", model: "gpt-5.4" }, houseStyle: true }));
return (h.connection.db.prepare("select compiled_prompt from skill_runtime_snapshots where id = ?").get(snapshotId) as { compiled_prompt: string }).compiled_prompt;
};
@@ -346,7 +353,7 @@ describe("built-in skills", () => {
modelProviderId: null, modelId: null, budgetCents: 1000, managerAgentId: null,
});
});
const snapshotId = h.connection.transaction(() => h.skills.resolveRuntimeSnapshot({ orgId: h.orgId, agentId, runId: "run_tiny", maxTokens: 10, taskText: "", runModel: { providerKey: "openai", model: "gpt-5.4" } }));
const snapshotId = h.connection.transaction(() => h.skills.resolveRuntimeSnapshot({ orgId: h.orgId, agentId, runId: "run_tiny", maxTokens: 10, taskText: "", runModel: { providerKey: "openai", model: "gpt-5.4" }, houseStyle: true }));
const prompt = (h.connection.db.prepare("select compiled_prompt from skill_runtime_snapshots where id = ?").get(snapshotId) as { compiled_prompt: string }).compiled_prompt;
expect(prompt).not.toContain("# Skill:");
expect(prompt).toContain("# Skills omitted for size");
@@ -373,7 +380,7 @@ describe("built-in skills", () => {
bindSkillSlugs({ db: h.connection.db, organization: h.organization, chat: new ChatService(h.connection.db) }, h.orgId, agentId, ["marketing-growth-hacker", "engineering-backend-architect"]);
});
const compile = (runId: string, maxTokens: number, taskText: string) => {
const snapshotId = h.connection.transaction(() => h.skills.resolveRuntimeSnapshot({ orgId: h.orgId, agentId, runId, maxTokens, taskText, runModel: { providerKey: "openai", model: "gpt-5.4" } }));
const snapshotId = h.connection.transaction(() => h.skills.resolveRuntimeSnapshot({ orgId: h.orgId, agentId, runId, maxTokens, taskText, runModel: { providerKey: "openai", model: "gpt-5.4" }, houseStyle: true }));
return (h.connection.db.prepare("select compiled_prompt from skill_runtime_snapshots where id = ?").get(snapshotId) as { compiled_prompt: string }).compiled_prompt;
};
const parts = compile("run_full", 1_000_000, "").split("\n\n---\n\n");
tests/unit/domain-services.test.ts
+4 -4
View File
@@ -24,7 +24,7 @@ describe("domain services", () => {
const connection = createConnection();
const rows = connection.db.prepare("select version from schema_migrations order by version asc").all() as Array<{ version: number }>;
expect(rows.map((row) => row.version)).toEqual([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65]);
expect(rows.map((row) => row.version)).toEqual([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66]);
connection.close();
});
@@ -186,7 +186,7 @@ describe("domain services", () => {
skills.bind({ orgId, agentId, skillId: skill.id, revisionId });
rmSync(skillDir, { recursive: true, force: true });
skills.runTests({ orgId, revisionId });
skills.resolveRuntimeSnapshot({ orgId, agentId, runId: "run-review-1", maxTokens: 44_800, taskText: "", runModel: { providerKey: "openai", model: "gpt-5.4" } });
skills.resolveRuntimeSnapshot({ orgId, agentId, runId: "run-review-1", maxTokens: 44_800, taskText: "", runModel: { providerKey: "openai", model: "gpt-5.4" }, houseStyle: true });
});
// Catalogue metadata ships in the snapshot; skill internals stay in the
@@ -247,10 +247,10 @@ describe("domain services", () => {
.get(orgA, "review-routes") as { id: string }).id;
expect(() => skills.bind({ orgId: orgA, agentId: agentB, skillId, revisionId })).toThrow("Agent was not found.");
skills.bind({ orgId: orgA, agentId: agentA, skillId, revisionId });
const firstResolve = skills.resolveRuntimeSnapshot({ orgId: orgA, agentId: agentA, runId: "run-review-1", maxTokens: 44_800, taskText: "", runModel: { providerKey: "openai", model: "gpt-5.4" } });
const firstResolve = skills.resolveRuntimeSnapshot({ orgId: orgA, agentId: agentA, runId: "run-review-1", maxTokens: 44_800, taskText: "", runModel: { providerKey: "openai", model: "gpt-5.4" }, houseStyle: true });
// Idempotent per run: a recovered run reuses its frozen snapshot
// instead of failing on the uniqueness guard.
expect(skills.resolveRuntimeSnapshot({ orgId: orgA, agentId: agentA, runId: "run-review-1", maxTokens: 44_800, taskText: "", runModel: { providerKey: "openai", model: "gpt-5.4" } })).toBe(firstResolve);
expect(skills.resolveRuntimeSnapshot({ orgId: orgA, agentId: agentA, runId: "run-review-1", maxTokens: 44_800, taskText: "", runModel: { providerKey: "openai", model: "gpt-5.4" }, houseStyle: true })).toBe(firstResolve);
expect(connection.db.prepare("select count(*) as n from skill_runtime_snapshots where run_id = 'run-review-1'").get())
.toEqual({ n: 1 });
});
tests/unit/skill-compatibility.test.ts
+1 -1
View File
@@ -42,7 +42,7 @@ describe("skill compatibility", () => {
skills.bind({ orgId, agentId, skillId: skill.id, revisionId });
});
const prompt = (runId: string, providerKey: string, model: string) => {
const id = connection.transaction(() => skills.resolveRuntimeSnapshot({ orgId, agentId, runId, maxTokens: 40_000, taskText: "", runModel: { providerKey, model } }));
const id = connection.transaction(() => skills.resolveRuntimeSnapshot({ orgId, agentId, runId, maxTokens: 40_000, taskText: "", runModel: { providerKey, model }, houseStyle: true }));
return connection.db.prepare("select compiled_prompt, source_revision_ids from skill_runtime_snapshots where id = ?").get(id) as { compiled_prompt: string; source_revision_ids: string };
};
tests/unit/skill-grants.test.ts
+26 -10
View File
@@ -7,6 +7,7 @@ import { OrganizationService } from "../../src/main/services/organization-servic
import { ChatService } from "../../src/main/services/chat-service";
import { SkillService } from "../../src/main/services/skill-service";
import { AGENT_TOOLS, dispatchAgentTool, type AgentToolContext } from "../../src/main/services/agent-tools";
import { writingModeLines } from "../../src/main/services/task-prompts";
function call(name: string, args: Record<string, unknown>) {
return { id: `call_${randomUUID()}`, name, arguments: args };
@@ -75,6 +76,21 @@ function importSkill(h: Harness, name: string, slug: string): void {
}
describe("skill grants", () => {
test("create_task records the writing mode, which decides house style and the writing rule", () => {
const h = harness();
const created = dispatchAgentTool(h.deps, h.ceoCtx, call("create_task", { title: "API spec", description: "Write the spec.", writing_mode: "technical" }));
const taskId = /task_id=(tsk_\w+)/.exec(created)![1];
expect(h.connection.db.prepare("select writing_mode from tasks where id = ?").get(taskId)).toEqual({ writing_mode: "technical" });
const plain = dispatchAgentTool(h.deps, h.ceoCtx, call("create_task", { title: "Chat", description: "Say hi." }));
expect(h.connection.db.prepare("select writing_mode from tasks where id = ?").get(/task_id=(tsk_\w+)/.exec(plain)![1])).toEqual({ writing_mode: "conversational" });
expect(dispatchAgentTool(h.deps, h.ceoCtx, call("create_task", { title: "X", description: "Y", writing_mode: "poetic" })))
.toBe('ERROR: "writing_mode" must be one of: conversational, technical, structured.');
expect(writingModeLines("conversational")).toEqual([]);
expect(writingModeLines("technical").join(" ")).toContain("Technical writing");
expect(writingModeLines("structured").join(" ")).toContain("Structured output");
h.connection.close();
});
test("the CEO runs only with granted skills, as the tool text says, and can grant itself", () => {
const h = harness();
importSkill(h, "Release Gate", "release-gate");
@@ -82,10 +98,10 @@ describe("skill grants", () => {
expect(listSkills?.description).not.toContain("every skill automatically");
expect(listSkills?.description).toContain("the CEO included");
const skills = new SkillService(h.connection.db, "");
expect(skills.hasRuntimeSkills(h.orgId, h.ceoId)).toBe(false);
expect(skills.hasRuntimeSkills(h.orgId, h.ceoId, true)).toBe(false);
expect(dispatchAgentTool(h.deps, h.ceoCtx, call("grant_skill", { agent_name: "CEO", skill: "release-gate" })))
.toBe("CEO now has the Release Gate skill; it joins their next run.");
expect(skills.hasRuntimeSkills(h.orgId, h.ceoId)).toBe(true);
expect(skills.hasRuntimeSkills(h.orgId, h.ceoId, true)).toBe(true);
h.connection.close();
});
@@ -100,11 +116,11 @@ describe("skill grants", () => {
expect(listing).toContain("Release Gate (release-gate): granted to Dana");
const skills = new SkillService(h.connection.db, "");
expect(skills.hasRuntimeSkills(h.orgId, h.danaId)).toBe(true);
expect(skills.hasRuntimeSkills(h.orgId, h.danaId, true)).toBe(true);
const revoked = dispatchAgentTool(h.deps, h.ceoCtx, call("revoke_skill", { agent_name: "Dana", skill: "release-gate" }));
expect(revoked).toContain("no longer has the Release Gate skill");
expect(skills.hasRuntimeSkills(h.orgId, h.danaId)).toBe(false);
expect(skills.hasRuntimeSkills(h.orgId, h.danaId, true)).toBe(false);
});
test("staff cannot hand out skills", () => {
@@ -112,7 +128,7 @@ describe("skill grants", () => {
importSkill(h, "Release Gate", "release-gate");
const denied = dispatchAgentTool(h.deps, h.danaCtx, call("grant_skill", { agent_name: "Dana", skill: "release-gate" }));
expect(denied).toContain("ERROR");
expect(new SkillService(h.connection.db, "").hasRuntimeSkills(h.orgId, h.danaId)).toBe(false);
expect(new SkillService(h.connection.db, "").hasRuntimeSkills(h.orgId, h.danaId, true)).toBe(false);
});
test("the humanizer house-style skill reaches every agent without bindings and cannot be revoked", () => {
@@ -120,18 +136,18 @@ describe("skill grants", () => {
const skills = new SkillService(h.connection.db, "");
// A non-forced skill grants nothing on its own - not even to the CEO.
importSkill(h, "Release Gate", "release-gate");
expect(skills.hasRuntimeSkills(h.orgId, h.ceoId)).toBe(false);
expect(skills.hasRuntimeSkills(h.orgId, h.danaId)).toBe(false);
expect(skills.hasRuntimeSkills(h.orgId, h.ceoId, true)).toBe(false);
expect(skills.hasRuntimeSkills(h.orgId, h.danaId, true)).toBe(false);
// Once the forced skill exists in the catalogue, everyone carries it.
importSkill(h, "humanizer", "humanizer");
expect(skills.hasRuntimeSkills(h.orgId, h.ceoId)).toBe(true);
expect(skills.hasRuntimeSkills(h.orgId, h.danaId)).toBe(true);
expect(skills.hasRuntimeSkills(h.orgId, h.ceoId, true)).toBe(true);
expect(skills.hasRuntimeSkills(h.orgId, h.danaId, true)).toBe(true);
// House style is not revocable, and granting it is a no-op with feedback.
const revoked = dispatchAgentTool(h.deps, h.ceoCtx, call("revoke_skill", { agent_name: "Dana", skill: "humanizer" }));
expect(revoked).toContain("ERROR");
expect(revoked).toContain("house style");
const granted = dispatchAgentTool(h.deps, h.ceoCtx, call("grant_skill", { agent_name: "Dana", skill: "humanizer" }));
expect(granted).toContain("automatically");
expect(skills.hasRuntimeSkills(h.orgId, h.danaId)).toBe(true);
expect(skills.hasRuntimeSkills(h.orgId, h.danaId, true)).toBe(true);
});
});