Hooks Reference

PreToolUse

.env File Protection

Blocks any Read, Grep, or Bash tool call whose input contains a path matching *.env*.

Property	Value	Notes
Event	PreToolUse	Runs before the tool is invoked
Matcher	Read|Grep|Bash	Pipe-separated tool names
Decision	deny	Stops execution and returns an error message
Reason	.env files are protected and cannot be read.	Shown to the user when blocked

The hook inspects the following fields from the tool input JSON:

• file_path (Read)

• path (Grep)

• glob (Grep)

• command (Bash)

A call is blocked if any of those values contains the string .env.

Complementary deny rules

The declarative permissions.deny rules in .claude/settings.json enforce the same restrictions at the tool-dispatch level, before the hook runs:

{
  "permissions": {
    "deny": [
      "Read(**/.env*)",
      "Grep(**/.env*)",
      "Bash(cat */.env*)",
      "Bash(cat .env*)"
    ]
  }
}

Both layers are necessary. See Quickstart for setup instructions.

UserPromptSubmit

/develop Suggestion

Fires before Claude processes each user message. If the prompt describes a development task, injects an additionalContext message recommending the /develop skill.

Property	Value	Notes
Event	UserPromptSubmit	Runs before Claude responds to the user’s message
Matcher	(none)	Fires on every prompt; filtering is done inside the script
Script	~/.claude/plugins/marketplaces/bytecode-solutions/hooks/suggest_develop.py	Resolved via the marketplace installation path; source lives in hooks/suggest_develop.py

Detection logic

The hook matches prompts containing development-intent keywords: implement, build, develop, feature, module, component, service, class, function, endpoint, api, refactor, create, new, and add <x> to/for/in.

Prompts that belong to other skills are excluded to avoid double-suggesting: run tests, run linters, run coverage, run security, check tests/linters/coverage/security/vulnerability, lint, audit, scan.

Behaviour

• If a match is found, outputs a JSON additionalContext message that Claude reads before responding, a nudge to recommend /develop, not a forced invocation.

• All errors are silently suppressed (contextlib.suppress) so the hook never blocks the user’s prompt.