Gemini Module

Optimization and portability guide for Forge modules. Ensures modules run seamlessly in Gemini CLI while maintaining full compatibility with Claude Code and OpenCode.

Portability Patterns

1. AI-Agnostic Module Resolution

Never rely on tool-specific environment variables alone. Use FORGE_MODULE_ROOT as the primary standard, with fallbacks.

In Bash Hooks:

MODULE_ROOT="${FORGE_MODULE_ROOT:-${CLAUDE_PLUGIN_ROOT:-$(builtin cd "$(dirname "$0")/.." && pwd)}}"
export FORGE_MODULE_ROOT="$MODULE_ROOT"

In SKILL.md (DCI calls):

dispatch skill-load <module>

In SKILL.md (body bash snippets): Do NOT use ${} variable expansion — Claude Code's permission system blocks it. Use hardcoded relative paths:

MODULE="Modules/<module-name>"
[ -d "$MODULE" ] || MODULE="."

2. Robust Binary Interface

Use the clap crate (Rust) to support standard CLI arguments. This allows Gemini to invoke tools directly with flags rather than relying solely on stdin JSON pipes.

• •Support --cwd for context steering.
• •Support --transcript-path for session analysis.
• •Provide --help for auto-discovery by AI agents.

Semantic Transcript Analysis

When building reflection or auditing modules, go beyond count-based heuristics.

1. •Extract Topics: Use regex to capture titles or topics following markers (e.g., ★ Insight: (.*)).
2. •Normalize and Match: slugify transcript topics and compare them against memory filenames.
3. •Avoid Date Prefixes: Use simple Title.md formats to make matching more reliable across different AI behaviors.

Gemini Specifics

• •Output Formatting: Use моноширинный-friendly Markdown.
• •Unicode Support: Gemini correctly renders unicode markers (★, ─, 📌). Use them for visual hierarchy in digests.
• •Skill Loading: Ensure skill-load.sh is invoked to inject project-specific steering into the session context.