fix(humanize2): pass agentDefaults to AgentRunCoordinator and support per-agent permission/sandbox/extraArgs

[zevorn]

Summary

Two fixes for humanize2 workflow agent launches that were blocking gen-idea and other workflows from completing successfully.

Bug 1: agentDefaults not passed to AgentRunCoordinator

Root cause: AgentRunCoordinator was constructed without agentDefaults in hub-server.ts. The config was loaded and passed to createHubHttpServer, but the AgentRunCoordinator (which actually creates and executes agent runs) never received it. This meant permissionMode, sandbox, and extraArgs from ~/.h2/config.yaml were silently ignored.

Impact:

• Claude agents ran with permissionMode: default, blocking MCP tool calls (artifact_deliver, artifact_get) that require user approval in headless claude -p mode
• Codex agents failed with "Not inside a trusted directory" when the working directory is not a git repo

Feature: Extended agent config surface

AgentModelDefaults now supports three new per-agent fields:

Field	Type	Purpose
permissionMode	PermissionMode	Claude CLI --permission-mode (e.g., bypassPermissions)
sandbox	SandboxMode	Codex CLI --sandbox
extraArgs	string[]	Arbitrary extra CLI args (e.g., --skip-git-repo-check, --dangerously-bypass-approvals-and-sandbox)

YAML parser supports both inline comma-separated values and multi-line list items for extraArgs.

Configuration example

agents:
  claude:
    model: claude-opus-4-7
    reasoningEffort: xhigh
    permissionMode: bypassPermissions
  codex:
    model: gpt-5.5
    reasoningEffort: xhigh
    extraArgs:
      - --dangerously-bypass-approvals-and-sandbox
      - --skip-git-repo-check

Changes

File	Change
src/config.ts	Extended AgentModelDefaults interface + YAML parser
src/hub/runs.ts	Applied permissionMode/sandbox/extraArgs from config defaults in createRun
src/hub-server.ts	Passed agentDefaults to AgentRunCoordinator constructor (critical bugfix)

Verification

Tested with gen-idea workflow:

• Before: 3/3 Codex explorers failed, Claude explorers needed per-tool permission allowlisting
• After: All 6 explorers completed successfully, idea-draft delivered correctly

Closes #165

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b95c3691ef

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[zevorn]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Another round soon, please!

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: bd0d0e6dce

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Reset agent section after parsing extraArgs list

After extraArgs: switches the parser into agents.extraArgs, the next agent header at indent 2 only updates currentAgent and leaves section unchanged. That means later fields (model, reasoningEffort, permissionMode, sandbox) are skipped because their parser branch requires section === "agents". In practice, a config with multiline codex.extraArgs followed by claude.model will silently ignore Claude overrides.

Useful? React with 👍 / 👎.