Factory Droid (factoryai-droid) hooks fail with schema mismatch on PreToolUse and PostToolUse — severely impacts Mission workflow

[dobedobedu]

What happened?

Entire's factoryai-droid hooks for PreToolUse and PostToolUse fail with exit code 1 on every invocation. This produces noisy error output in the Droid CLI on every tool call while Entire receives no data.

PreToolUse error:

HOOKS  PreToolUse → Task
 └─ Script: entire hooks factoryai-droid pre-tool-use
 └─ Result : Exit code: 1
   stderr:
   failed to capture pre-task state: tool_use_id is required

PostToolUse error:

HOOKS  PostToolUse → Task
 └─ Script: entire hooks factoryai-droid post-tool-use
 └─ Result : Exit code: 1
   stderr:
   failed to parse hook event: failed to parse hook input: json: cannot unmarshal string into Go struct field postToolHookInputRaw.tool_response of type struct { AgentID string "json:\"agentId\"" }

Impact on Droid Missions

This is especially painful during Droid Missions — Factory's orchestrated multi-subagent workflow feature. Missions spawn multiple parallel Task (subagent) tool calls, and each one triggers the failing PreToolUse hook. With a mission running 5-10+ subagents, this means:

• 10-20+ failing hook executions per mission cycle
• Each failing hook adds latency to the already-parallelized workflow
• The cumulative slowdown dramatically degrades the automated/agentic experience
• The constant error output makes it hard to follow what the mission is actually doing

For context, a typical mission cycle looks like:

HOOKS  Stop
 └─ Script: entire hooks factoryai-droid stop
 └─ Result : Exit code: 0

HOOKS  PreToolUse → Task
 └─ Script: entire hooks factoryai-droid pre-tool-use
 └─ Result : Exit code: 1    ← FAILS

HOOKS  PreToolUse → Task
 └─ Script: entire hooks factoryai-droid pre-tool-use
 └─ Result : Exit code: 1    ← FAILS

HOOKS  PreToolUse → Task
 └─ Script: entire hooks factoryai-droid pre-tool-use
 └─ Result : Exit code: 1    ← FAILS
... (repeats for every subagent)

Root cause

Two schema mismatches between what Entire expects and what Factory Droid provides in hook input JSON:

1. Missing tool_use_id: Entire's pre-tool-use handler requires a tool_use_id field, but Droid's PreToolUse hook input does not include it. This is the same issue documented for Claude Code at PostToolUse hook not sending tool_use_id field as documented anthropics/claude-code#13241 — tool_use_id was never reliably sent by the agent platform.

2. tool_response type mismatch: Entire expects tool_response to be a structured object with an agentId field (struct { AgentID string }), but Droid sends tool_response as a plain string (the raw tool output text). This causes the Go JSON unmarshal error.

Droid's actual hook input format (v0.98.0)

Per Factory docs (https://docs.factory.ai/reference/hooks-reference), Droid sends:

PreToolUse:

{
  "session_id": "...",
  "transcript_path": "...",
  "cwd": "...",
  "permission_mode": "...",
  "hook_event_name": "PreToolUse",
  "tool_name": "Task",
  "tool_input": { ... }
}

No tool_use_id field.

PostToolUse:

{
  "session_id": "...",
  "transcript_path": "...",
  "cwd": "...",
  "permission_mode": "...",
  "hook_event_name": "PostToolUse",
  "tool_name": "Task",
  "tool_input": { ... },
  "tool_response": "<raw string output, not a structured object>"
}

tool_response is a string, not an object with agentId.

Steps to reproduce

1. Install Entire CLI v0.5.3
2. In a git repo, run entire enable --agent factoryai-droid
3. Start Droid in that repo: droid
4. Ask Droid to perform any tool call (e.g. read a file)
5. Observe PreToolUse and PostToolUse hook errors in Droid output
6. Run a Droid Mission (multi-subagent workflow) to see the amplified impact

Entire CLI version

0.5.3

OS and architecture

macOS ARM64 (darwin/arm64)

Agent

Factory Droid (factoryai-droid)

Suggested fix

1. Make tool_use_id optional in the PreToolUse parser — generate a fallback ID or skip the field when absent
2. Handle tool_response as either a string or a structured object (union type / json.RawMessage) rather than requiring the agentId struct
3. Alternatively, correlate tool calls via tool_name + session_id + timestamp instead of requiring tool_use_id
4. Consider graceful degradation: if a hook fails to parse, continue with partial data rather than exiting with code 1 on every invocation

Additional context

• Related: Entire startup hooks are failing #237 (hooks failing with Claude Code for similar schema reasons)
• Related: Summary generation fails: cannot unmarshal array from Claude Code 2.x JSON output #750 (cannot unmarshal Go struct — same class of JSON parsing error)
• Related: Factory Droid checkpoints show 'Unknown' agent in entire.io even though checkpoint metadata contains the correct agent/model #859 (Droid agent shows as 'Unknown' — may be downstream of hooks failing to capture data)
• Anthropic-side root cause: PostToolUse hook not sending tool_use_id field as documented anthropics/claude-code#13241 (tool_use_id not sent by agent platforms)
• Droid hook docs: https://docs.factory.ai/reference/hooks-reference
• Workaround: Remove all Entire hooks from .factory/settings.json project settings (which is what I have had to do to make Missions usable)