✔ sessionModel: agent vars win, then the system's execution settings (14.470615ms) ✔ a pi bundle: prompt, policy, tools and manifest, 0600 in a 0700 directory (5.105281ms) ✔ a chief-of-staff bundle lists launch among the bus tools and the manifest's typed tools (2.782257ms) ✔ a chief-of-staff bundle lists end_run and change_organization as typed and only instance.name untyped, for either harness (4.343517ms) ✔ the stop line names end_run for a role.end holder, and every other role keeps today's text (6.82277ms) ✔ a claude-code bundle adds the wrapped gate hook and the MCP config (2.118636ms) ✔ the prompt's built-in tools line is the harness's own names, as the manifest lists them (8.999916ms) ✔ the prompt names the project clones in the workspace, and none when there are none (7.022935ms) ✔ the prompt explains the review loop to a role with push_working, and reviews to one with fetch_review (9.2354ms) ✔ the prompt explains landing to a role with land_work, and to no other (5.608538ms) ✔ the prompt tells the pm to ask the chief of staff for sessions, and no other role (9.686157ms) ✔ a bundle is written once: an existing file refuses (1.688149ms) ✔ a path with a single quote can't go into the hook command (1.768643ms) ✔ web grant on pi: the bundle has web_search and web_fetch, and no built-in web tool (2.355228ms) ✔ no web grant on pi: no web tool in the bundle, and the gate blocks a web call (1.806924ms) ✔ web grant on claude-code: the bundle has web_search and web_fetch, and no built-in web tool (2.12424ms) ✔ no web grant on claude-code: no web tool in the bundle, and the gate blocks a web call (1.768354ms) ✔ a granted session's gate lets the typed web tools through and still blocks the built-ins (1.762274ms) ✔ search-source searxng-url: a granted role gets the typed broker tools and no harness web tool; an ungranted role gets nothing (5.431495ms) ✔ search-source searxng-bundled: a granted role gets the typed broker tools and no harness web tool; an ungranted role gets nothing (5.25466ms) ✔ search-source harness-builtin: claude-code gets WebSearch and WebFetch and no typed web tool, only when granted (9.244148ms) ✔ search-source harness-builtin is refused for a granted pi role, which has no built-in web tool (1.874038ms) ✔ allow exits 0, a deny exits 2 with the reason on stderr (148.614026ms) ✔ a missing or wrong policy, or a bad event, exits 2 (88.27362ms) ✔ the bundle's wrapped command: a missing gate or node still blocks (1105.642476ms) ✔ claude: typed tools through MCP, the hook blocks, builtins outside --tools don't exist (852.535753ms) ✔ claude: the hook alone blocks a path outside the workspace (541.491097ms) ✔ claude: a second turn resumes the first turn's session (861.443578ms) ✔ claude adapter: --restricted is always passed (5.562868ms) ✔ claude adapter: MOSAIC_CLAUDE_BIN wins over PATH and must be an absolute executable (19.02818ms) ✔ claude adapter: a directory pin exits 2 before it runs; a pin with a space runs (5.898221ms) ✔ claude: CLAUDE.md files and auto-memory don't reach the model; without --restricted they do (837.990358ms) ✔ claude: a missing hook or MCP file refuses before claude starts (7.498473ms) ✔ pi: policy tools and typed tools pass, anything else is blocked (3.986637ms) ✔ claude: builtins map from pi names, typed tools need the mcp prefix (0.848426ms) ✔ file tool paths must resolve inside the workspace (1.689154ms) ✔ pi's own path normalisation can't be used to step out (1.096153ms) ✔ a symlink inside the workspace that points out is outside (0.76199ms) ✔ a dangling symlink is refused at any depth, in both harnesses (4.653366ms) ✔ read is checked under every spelling pi's read would open, in both harnesses (16.547904ms) ✔ other spellings cover directories, dangling links and pi's cwd (1.478243ms) ✔ a relative path climbs from the workspace's real path, in both harnesses (5.617836ms) ✔ claude path fields per tool (0.806799ms) ✔ glob patterns stay inside the workspace (0.984971ms) ✔ a path that can't be checked is blocked (0.575386ms) ✔ fetch_review's have lists the working refs before origin's, and no more than HAVE_MAX (254.486544ms) ✔ fetch_review's have orders the working refs newest first within each clone, one clone after another (156.367153ms) ✔ a runner git past its time: SIGTERM reaches its process group, and SIGKILL follows after the grace period (1810.5941ms) ✔ once a signalled runner git has closed, its group gets SIGKILL, so a process that ignored SIGTERM dies too (304.200473ms) ✔ land_work sends one land line and resolves the host's started answer (3.596731ms) ✔ initialize, ping and tools/list (48.336886ms) ✔ tools/call goes through the tool socket; a refusal is an isError result (38.575658ms) ✔ replies are compact JSON; over the cap a tool error names the size and the cap (41.136237ms) ✔ unknown tools and methods are JSON-RPC errors and never reach the socket (36.015494ms) ✔ a missing argument is a usage error (29.470574ms) ✔ pi: typed tools reach the socket, the gate blocks, agent_end writes the marker (406.754349ms) ✔ pi: a reply over the cap is a tool error naming the size and the cap; list_tasks asks for open tasks (332.266257ms) ✔ pi: a write through a dangling symlink is blocked, and nothing appears outside (353.220521ms) ✔ pi: a read is refused when pi would open another spelling outside (379.942733ms) ✔ pi: a relative path climbs from the real path of a workspace behind a symlink (346.474997ms) ✔ pi: a relative path climbs from the real path of a dataRoot behind a symlink (350.780815ms) ✔ pi: a missing extension refuses before any model call (6.550712ms) ✔ pi: an extension without its configuration fails pi's start (271.284323ms) ✔ founderCheck: founder variables, then a needed service without a usable token (1.677902ms) ✔ turnRequest names the sender, class, reply and decision (0.280373ms) ✔ turnRequest says a tester's message is a test and never Jason's instruction (0.283654ms) ✔ turnRequest shows the message's own sent time, a day old as well as new (0.213402ms) ✔ stateBlock names the launch state with its time and event, and the open tasks with the cap (0.330713ms) ✔ a message becomes a turn, the answer goes back as a RESULT, SIGTERM releases and exits 0 (290.613198ms) ✔ launches revoked and then restored: the turn's block reads allowed and gives the restore time (505.917796ms) ✔ a task closed on the tracker is missing from the next turn's open list (286.371507ms) ✔ a message read a day after it was sent shows the time it was sent (199.710819ms) ✔ a decision the human resolves becomes a turn citing the decision and the answer (245.896563ms) ✔ a failed state read fails the turn: no adapter runs, the sender hears why, the message is read (182.891099ms) ✔ a SIGTERM before the claim stops the runner with exit 0 and no claim (103.912862ms) ✔ typed tools carry the runner's capability; launch goes to the host's launch socket (446.473219ms) ✔ end_run and change_organization reach the broker with the tool's arguments, and refusals keep their codes (479.404431ms) ✔ a RESULT gets no automatic reply; failed turns reply with the reason (1321.839287ms) ✔ the bus's launch notice is one turn with no reply, and is not taken again (261.775291ms) ✔ SIGTERM during a turn kills the turn's process group and still exits 0 (173.938243ms) ✔ founder credentials stop before the claim (20) (198.297001ms) ✔ a refused claim exits 21; an ended run's capability exits 22 (220.360623ms) ✔ the launch ending under a running session exits 22 (160.449572ms) ✔ a broker that stays unreachable exits 23 after brokerRetries polls (261.339622ms) ✔ a broker that is down at the claim exits 23, not 21 (91.357477ms) ✔ no capability, or a malformed one, on stdin exits 2 (202.863757ms) ✔ a missing or malformed policy exits 2 before the claim (126.446362ms) ✔ every role sees read_spec, record writer follows its grant, citation schemas stay strict (14.975904ms) ✔ typed socket forwards read content/history and agent record writes, and cannot write a spec (244.965832ms) ✔ the PM gets its task verbs and the reads, and no launch (10.073706ms) ✔ the chief of staff gets launch, send_message, end_run and change_organization, and no task verbs (3.413148ms) ✔ request_staff is the pm's alone: staff.request over message.send, roles and counts bounded (11.840048ms) ✔ end_run sends role and reason, no decision; change_organization sends revision and the organization keys (2.408321ms) ✔ the organization keys change_organization takes are the broker's ORGANIZATION_KEYS (0.169071ms) ✔ a role without role.end or organization.change gets neither tool, and each tool follows its own grant (8.724041ms) ✔ the assistant gets send_message, the reads and the web tools (it holds the web grant), no launch (1.415957ms) ✔ list_requirements reads the requirement ids, for a task.create holder with a tracker (2.644465ms) ✔ a coder gets no launch, no resolve_decision, and no task tools without a tracker (1.346908ms) ✔ a coder with a tracker gets its own task tools and no task_reassign (1.138529ms) ✔ launch only when the business's launch block names the instance as launcher (2.489388ms) ✔ an action outside the instance's authority has no tool (1.416605ms) ✔ the review loop: push_working and request_review for the coder, fetch_review and review_verdict for the reviewer, none for the PM (5.189034ms) ✔ land_work is there only with land, after push_working, with the task and the decision (3.01795ms) ✔ callTool: one JSON line out, the result back, a refusal rejects (11.597736ms) ✔ list_tasks takes the broker's filters and pages; get_task takes a task_ref (2.451633ms) ✔ callTool sends list_tasks done false unless the session names done; other tools as given (2.597917ms) ✔ replyText: compact JSON up to the cap, an error naming the size and the cap over it (0.599374ms) ✔ web_search's query maxLength is the web layer's SEARCH_MAX_QUERY_CHARS (2.51049ms) ℹ tests 108 ℹ suites 0 ℹ pass 108 ℹ fail 0 ℹ cancelled 0 ℹ skipped 0 ℹ todo 0 ℹ duration_ms 12618.74532