[plan] Expand language ecosystem inference for network.allowed in create-agentic-workflow prompt

[github-actions]

Objective

Improve language/ecosystem inference in the create-agentic-workflow prompt so the agent automatically adds the correct package registry domains to network.allowed based on detected stack indicators, without requiring explicit user prompting.

Context

From Agent Persona Exploration discussion #18250, the agent sometimes needs explicit prompting to add the right ecosystem to network.allowed (e.g., registry.npmjs.org for Node, pypi.org for Python, proxy.golang.org for Go). It relies on the user mentioning the stack rather than inferring it from context clues.

Approach

1. Locate the create-agentic-workflow prompt/skill
2. Expand the language indicator list with explicit mappings:
   • Node.js / npm: detect package.json, .nvmrc, node_modules → registry.npmjs.org
   • Python / pip: detect requirements.txt, pyproject.toml, setup.py → pypi.org, files.pythonhosted.org
   • Go: detect go.mod, go.sum → proxy.golang.org, sum.golang.org
   • Ruby: detect Gemfile → rubygems.org
3. Add prompt guidance: "If running tests or installing packages, infer the ecosystem from the repository's language files and add the appropriate registries to network.allowed"
4. This complements the existing network permission guidance

Files to Modify

• Update: the create-agentic-workflow prompt/skill with expanded language inference indicators

Acceptance Criteria

• Prompt includes explicit ecosystem → network domain mappings
• Agent infers package registry from repo language indicators without user prompting
• Agent correctly suggests network.allowed entries for all major ecosystems (Node, Python, Go, Ruby)

Generated by Plan Command for issue #discussion #18250

• expires on Feb 27, 2026, 4:46 AM UTC