Kiro

Install methods

• Not wired: agentseam records this agent but has no adapter for its hooks, so chock has nothing to wire.

Evidence wanted

Chock has not seen this agent run its hooks yet. If you use it, you can help: agentseam issue #183

What it can enforce

Tier

unadapted

Grade before a tool runs

none

Fail mode before a tool runs

n/a

Basis

vendor-docs

Recorded

2026-10-11

Ask support

The engine records no ask claim for this agent.

Hook timeout

The engine records no timeout for this agent.

Test status

• No row in the engine's list of live tests for this agent.

Known limits

Unadapted, not none: Kiro has a real command-hook surface, but its dialect is one no agentseam family speaks. The config is a versioned list of named hooks with a trigger and an action object, which no install wrapper writes; the verdict is an exit code only, and the IDE and CLI read it differently (IDE: any non-zero exit blocks Pre Tool Use; CLI: only exit 2 blocks, other codes fail open); and the payload is Claude-shaped (hook_event_name, cwd, session_id, tool_name, tool_input) with no key that separates it from a Claude Code payload. Building it means a new family module, which is a core change outside a vendor-config batch. Value-based detection (D1) now declines a Kiro payload naming an MCP tool as mcp_<server>_<tool> (Claude Code spells it mcp__<server>__<tool>) or a tool alias Claude Code never sends (fs_read, fsRead, fs_write, fsWrite, execute_bash, execute_cmd, use_aws; kiro.dev/docs/reference/built-in-tools, updated 2026-10-10), so it detects as no agent. One naming a V3 primary tool (read, write, shell), or carrying no tool (SessionStart, UserPromptSubmit, Stop), has nothing that separates it and is still recognised as Claude Code, whose deny is a JSON body with exit 0; Kiro reads only the exit code, so that deny would not block. Kiro is not a supported target: do not wire agentseam into it by hand. D2a (2026-10-11) re-checked this and kept the row unadapted: D1's detection covers only part of Kiro's documented events (the alias and mcp_<server>_<tool> tool calls), not the V3 primary-tool or tool-less payloads, and an adapter cannot claim those without making Claude Code payloads ambiguous; the exit-code-only verdict and the versioned hook file would also need a new family. The docs also disagree on the timeout field (timeout in seconds, default 60, vs timeout_ms, default 30,000). Built-in tools are execute_bash (execute_pwsh on Windows) and fs_write, fs_append, str_replace, delete_file. Instruction files reach it through AGENTS.md. Evidence wanted (to test or overturn this row): https://github.com/open-coder-ai/agentseam/issues/183