astrid-runtime/astrid

Summary crates/astrid-hooks/src/handler/agent.rs contains a fully stubbed AgentHandler that: execute() always returns HookExecutionResult::Skipped is_available() always returns false Has an unused AgentConfig with a hardcoded "claude-3-haiku" model string The HookHandler::Agent variant exists in the enum but does nothing Why remove it Hooks are for deterministic, fast interception: shell scripts, HTTP webhooks, WASM modules. Routing a hook through an LLM adds latency, non-determinism, and cost to every intercepted event. If someone wants LLM-based decision making on events, that belongs in an orchestrator capsule, not the hook system. The stub was written speculatively and never implemented. It adds dead code surface area, a HookHandler variant that confuses the API, and an AgentFailed error variant in HandlerError that can never be returned. Changes Delete crates/astrid-hooks/src/handler/agent.rs Remove AgentHandler from handler/mod.rs Remove HookHandler::Agent { .. } variant Remov

Summary Found as an out-of-scope observation while reviewing PR for #375. CapsuleRegistry::register and CapsuleRegistry::register_connector both return CapsuleError::UnsupportedEntryPoint when a duplicate ID is detected, but their doc comments claim they return CapsuleError::AlreadyRegistered and CapsuleError::ConnectorAlreadyRegistered respectively — neither of which exists in the CapsuleError enum. Impact Callers cannot reliably match on duplicate-registration errors without fragile string inspection. Any code written against the (false) documented API will silently fall through to a wildcard arm. Error messages in logs read as "Unsupported entry point: Already registered: foo", which is misleading. PR #375 cleaned up the symmetric unregister / unregister_connector paths (introducing NotFound) but did not address the register pair. Files / Lines crates/astrid-capsule/src/registry.rs line 60 (doc comment for register) crates/astrid-capsule/src/registry.rs line 66-68 (returns Unsu

Bug Found as an out-of-scope finding during review of PR fixing issue #445. File: crates/astrid-events/src/ipc.rs Line: 731 Description The test is_known_tag_covers_all_variants sets EXPECTED_VARIANT_COUNT = 17 and has 17 representatives. However IpcPayload has 19 variants: RawJson, UserInput, AgentResponse, ApprovalRequired, ApprovalResponse, OnboardingRequired, LlmRequest, LlmStreamEvent, LlmResponse, ToolExecuteRequest, ToolExecuteResult, ToolCancelRequest, SelectionRequired, ElicitRequest, ElicitResponse, Connect, Disconnect, Custom, Unknown The test omits ToolCancelRequest from the representatives list entirely, so the test never verifies that is_known_tag correctly recognizes "tool_cancel_request" and that the variant serializes to a known tag. Unknown is a #[serde(other)] catch-all that is intentionally excluded (it has no type tag of its own). But ToolCancelRequest is a real publishable variant and should be represented. Impact If is_known_tag ever loses coverage of "tool_canc

Describe the bug In crates/astrid-cli/src/tui/render.rs, the markdown parser for numbered lists uses strip_prefix(|c: char| c.is_ascii_digit()). Since strip_prefix with a closure only matches a single character, it only consumes one digit. If a list item starts with 10. , it consumes 1, leaving 0. , and then .and_then(|s| s.strip_prefix(". ")) fails. To Reproduce Have the assistant output a markdown numbered list with 10 or more items. Observe that items 1-9 are styled correctly (indented with colored numbers). Observe that item 10 and beyond are rendered as plain unstyled text without indentation. Suggested Fix Change the parsing logic to consume all leading digits before checking for . . Note: Found as an out-of-scope finding during review of PR feat/multi-line-paste.

In crates/astrid-capsule/src/engine/mcp.rs, invoke_interceptor does not check tool_result.is_error. When astrid_hook_intercept returns isError: true, the Rust side returns Ok with the error text as payload content. Hook errors are silently treated as success. Fix: check is_error and return Ok with empty vec if true, with a warn log. Found as OOS finding during review of feat/openclaw-pipeline-integration.