fix(prompt): simplify system prompts, remove massive duplication (~1795 tokens saved)

[ranxianglei]

Summary

Two-part change: (1) eliminate massive prompt duplication saving ~1795 tokens, (2) add per-model prompt configurability via options.prompt.

Part 1: Prompt Simplification (~1795 tokens saved)

default.txt (GLM-5.x default prompt): 7638 → 3038 chars (~1150 tok saved)

Was a merge artifact — two versions concatenated end-to-end. 6 sections appeared twice (# Proactiveness, # Following conventions, # Code style, # Doing tasks, # Tool usage, # Code References), and 4 of 6 examples were exact duplicates of the first half. Merged into one clean version, preserving all unique instructions from both halves.

shell.txt (Bash tool description): 6287 → 5253 chars (~260 tok saved)

The parallel-tool-call instruction (You can call multiple tools in a single response. When multiple independent pieces...) was repeated 4 times verbatim (commit steps + PR steps). Deduplicated to concise Run the following ... in parallel:. Removed duplicate commit rule. All ${...} template variables preserved.

anthropic.txt (Claude prompt): 8212 → 6672 chars (~385 tok saved)

Two verbose Task Management examples (~900 chars each) demonstrating identical TodoWrite usage merged into one concise example. Fixed broken # Doing tasks section (orphan - on its own line).

Part 2: Per-Model Prompt Configurability

system.ts: options.prompt config override

Added a PROMPT_MAP and config override check at the top of provider(). Users can now specify which system prompt file a model uses via model config:

{
  "provider": {
    "glm": {
      "models": {
        "glm-5.2": { "options": { "prompt": "anthropic" } }
      }
    }
  }
}

Valid values: default, anthropic, beast, gemini, gpt, kimi, codex, trinity.

This enables:

• Protocol-based routing: point OpenAI-compatible models to the GPT prompt
• Intelligence-based routing: use leaner prompts for smarter models
• A/B testing: swap prompts per model without code changes

The hardcoded provider() routing remains as the auto-detect fallback.

Verification

• bun run typecheck — passes (all 18 packages)
• Pre-push hook (bun turbo typecheck) — 12/12 tasks successful
• All template variables (${...}) preserved in shell.txt
• No behavioral change — prompt trimming only removes redundancy; config override is opt-in

[github-actions]

Thanks for your contribution!

This PR doesn't have a linked issue. All PRs must reference an existing issue.

Please:

1. Open an issue describing the bug/feature (if one doesn't exist)
2. Add Fixes #<number> or Closes #<number> to this PR description

See CONTRIBUTING.md for details.

[github-actions]

This PR doesn't fully meet our contributing guidelines and PR template.

What needs to be fixed:

• PR description is missing required template sections. Please use the PR template.

Please edit this PR description to address the above within 2 hours, or it will be automatically closed.

If you believe this was flagged incorrectly, please let a maintainer know.

[github-actions]

This pull request has been automatically closed because it was not updated to meet our contributing guidelines within the 2-hour window.

Feel free to open a new pull request that follows our guidelines.