tony/CHORUS
1
0
Fork 0
Code Issues 10 Pull Requests Actions Packages Projects Releases Wiki Activity
Files
9190c75440dbdc23feab4249f105a19995de25a1
CHORUS/prompts
History
anthonyrawlins df5ec34b4f feat(execution): Add response parser for LLM artifact extraction 
Implements regex-based response parser to extract file creation actions
and artifacts from LLM text responses. Agents can now produce actual
work products (files, PRs) instead of just returning instructions.

Changes:
- pkg/ai/response_parser.go: New parser with 4 extraction patterns
  * Markdown code blocks with filename comments
  * Inline backtick filenames followed by "content:" and code blocks
  * File header notation (--- filename: ---)
  * Shell heredoc syntax (cat > file << EOF)

- pkg/execution/engine.go: Skip sandbox when SandboxType empty/none
  * Prevents Docker container errors during testing
  * Preserves artifacts from AI response without sandbox execution

- pkg/ai/{ollama,resetdata}.go: Integrate response parser
  * Both providers now parse LLM output for extractable artifacts
  * Fallback to task_analysis action if no artifacts found

- internal/runtime/agent_support.go: Fix AI provider initialization
  * Set DefaultProvider in RoleModelMapping (prevents "provider not found")

- prompts/defaults.md: Add Rule O for output format guidance
  * Instructs LLMs to format responses for artifact extraction
  * Provides examples and patterns for file creation/modification
  * Explains pipeline: extraction → workspace → tests → PR → review

Test results:
- Before: 0 artifacts, 0 files generated
- After: 2 artifacts extracted successfully from LLM response
- hello.sh (60 bytes) with correct shell script content

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
2025-10-11 22:08:08 +11:00
..
defaults.md
feat(execution): Add response parser for LLM artifact extraction
2025-10-11 22:08:08 +11:00
human-roles.yaml
Implement Phase 2 & 3: Complete HAP Terminal Interface with Patch Management
2025-09-07 09:38:14 +10:00
README.md
docs(prompts): add README with structure, schema, and usage
2025-09-06 15:46:17 +10:00
roles.yaml
feat(prompts): load system prompts and defaults from Docker volume; set runtime system prompt; add BACKBEAT standards
2025-09-06 15:42:41 +10:00

README.md

CHORUS Prompts Directory

This directory holds runtime‑loaded prompt sources for agents. Mount it into containers to configure prompts without rebuilding images.

  • Role prompts (S): YAML files defining agent roles and their system_prompt.
  • Default instructions (D): A shared Markdown/TXT file applied to all agents.
  • Composition: Final system prompt = S + two newlines + D.

Mounting and Env Vars

  • Bind mount (example): -v /srv/chorus/prompts:/etc/chorus/prompts:ro
  • Env vars:
    • CHORUS_PROMPTS_DIR=/etc/chorus/prompts
    • CHORUS_DEFAULT_INSTRUCTIONS_PATH=/etc/chorus/prompts/defaults.md (optional)
    • CHORUS_ROLE=<role-id> (selects which S to compose with D)

Reload: prompts are loaded at startup. Restart containers to pick up changes.

Files and Structure

  • defaults.md (or defaults.txt): global Default Instructions D
  • roles.yaml (optional): multiple roles in one file
  • *.yaml / *.yml: one or more files; all are merged by role ID

Example layout:

/etc/chorus/prompts/
  defaults.md
  roles.yaml
  ops.yaml
  analysts.yaml

Role YAML Schema

Top-level key: roles. Each role is keyed by its role ID (used with CHORUS_ROLE).

roles:
  arbiter:
    name: "Arbiter"
    description: "Coordination lead for cross-agent planning and consensus."
    tags: [coordination, planning]
    system_prompt: |
      You are Arbiter, a precise coordination lead...
    defaults:
      models: ["meta/llama-3.1-8b-instruct"]
      capabilities: ["coordination","planning","dependency-analysis"]
      expertise: []
      max_tasks: 3

Notes:

  • Role IDs must be unique across files; later files overwrite earlier definitions of the same ID.
  • system_prompt is required to meaningfully customize an agent.
  • defaults are advisory and can be used by orchestration; runtime currently composes only the prompt string (S + D).

Default Instructions (D)

Provide generic guidance for all agents, including:

  • When/how to use HMMM (collaborative reasoning), COOEE (coordination), UCXL (context), BACKBEAT (timing/phase telemetry)
  • JSON envelopes for each message type
  • Safety, auditability, and UCXL citation policy

File can be Markdown or plain text. Example: defaults.md.

Composition Logic

  • If CHORUS_ROLE is set and a matching role exists: final system prompt = role.system_prompt + "\n\n" + defaults
  • Else, if defaults present: final = defaults
  • Else: fallback to a minimal built‑in default

Validation Tips

  • Ensure mounted path and env vars are set in your compose/stack.
  • Role ID used in CHORUS_ROLE must exist in the merged set.
  • Restart the container after updating files.

BACKBEAT Standards

Include clear guidance in defaults.md on:

  • Phases: prepare|plan|exec|verify|publish
  • Events: start|heartbeat|complete
  • Correlation: team_id, session_id, operation_id, link to COOEE/HMMM IDs
  • Budgets/latency, error reporting, and UCXL citations for escalations

See the repository’s prompts/defaults.md for a complete example.