aea4d45fd8
🎭 Phase 2: HAP Terminal Interface Implementation ✅ **Core Terminal Interface**: Interactive command-driven HAP terminal with help system ✅ **HMMM Message Composition System**: - New reasoning messages, thread replies, network queries, decision proposals - Complete message metadata handling (topics, threads, timestamps) ✅ **UCXL Context Browsing System**: - Address parsing, content retrieval from DHT encrypted storage - Search functionality, content creation, history navigation ✅ **Decision Participation System**: - Active decision listing, decision details with voting status - Vote casting with reasoning, decision proposals, HMMM integration 🔧 Phase 3: Enhanced Human Workflows ✅ **Patch Creation and Submission Workflows**: - Complete patch lifecycle management (create, review, submit, track) - Multiple patch types (context, code, config, docs) - UCXL integration with DHT storage, HMMM coordination ✅ **Time-Travel Diff Support**: - Temporal navigation operators (~~<n>, ^^<n>, @<time>) - Decision-hop analysis, visual diff display, version comparison 🏗️ **Architecture Highlights**: - **Multi-binary structure**: Separate chorus-agent and chorus-hap binaries - **Shared P2P runtime**: Both binaries use identical libp2p, DHT, HMMM, UCXL systems - **Interactive sub-shells**: Dedicated command environments for HMMM, UCXL, patches, decisions - **Network integration**: All features connect to distributed P2P agent network - **Human-agent parity**: Humans participate as first-class network citizens 📦 **New Files**: - internal/hapui/terminal.go: Complete HAP terminal interface (2400+ lines) - prompts/human-roles.yaml: Role-based prompt configuration - docs/decisions/*: HAP conversion decision record 🔗 **Integration Points**: - HMMM: Collaborative reasoning and patch/decision announcements - UCXL: Context addressing and version management - DHT: Distributed storage of patches and content - Decision System: Formal approval and consensus workflows The HAP terminal interface now provides comprehensive human portal into the CHORUS autonomous agent network, enabling collaborative reasoning, context sharing, patch management, and distributed decision-making between humans and AI agents. 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
47 lines
2.4 KiB
Markdown
47 lines
2.4 KiB
Markdown
# Decision Record: Convert Human Markdown Prompts to CHORUS Role YAML
|
||
|
||
- Date: 2025-09-06
|
||
- UCXL Address: ucxl://arbiter:ops@CHORUS:prompt-migration/#/docs/decisions/2025-09-06-convert-human-prompts-to-roles-yaml.md
|
||
|
||
## Problem
|
||
Human-oriented prompt templates exist as Markdown files under `agentic-ai-prompt-templates/human/`. CHORUS now sources agent role prompts (S) and default instructions (D) at runtime from bind-mounted YAML/Markdown files. We need these human templates available in the new YAML format to configure agents via Docker volume binding without rebuilding images.
|
||
|
||
## Options Considered
|
||
1) Manual conversion of each Markdown file to a YAML role entry
|
||
- Pros: Tight editorial control
|
||
- Cons: Time-intensive, error-prone, hard to keep in sync
|
||
|
||
2) Automated converter script to parse Markdown sections and emit a consolidated `system_prompt` with metadata
|
||
- Pros: Fast, repeatable, easy to re-run when templates change
|
||
- Cons: Heuristics may miss atypical structures; requires review
|
||
|
||
3) Store raw Markdown and embed at runtime
|
||
- Pros: No conversion step
|
||
- Cons: Diverges from adopted loader schema, complicates composition and validation
|
||
|
||
## Decision
|
||
Adopt Option 2. Add a utility script `utilities/convert_human_prompts_to_yaml.py` that:
|
||
- Reads `agentic-ai-prompt-templates/human/*.md`
|
||
- Extracts title, Description, Tools, Use Cases, When to Use
|
||
- Constructs `system_prompt` as: "You are <Name>." + Description + Tools + Use Cases + When To Use
|
||
- Emits `project-queues/active/CHORUS/prompts/human-roles.yaml` with one role per file, using filename as role ID
|
||
- Populates advisory `defaults` (models/capabilities/expertise/max_tasks)
|
||
|
||
## Impact
|
||
- Roles become mountable via `CHORUS_PROMPTS_DIR` (e.g., `-v ../prompts:/etc/chorus/prompts:ro`)
|
||
- Agents can select any converted role via `CHORUS_ROLE=<role-id>`
|
||
- Future updates to human Markdown can be re-converted by re-running the script
|
||
|
||
## Rollback
|
||
- Remove `human-roles.yaml` from the prompts directory
|
||
- Agents will continue to use existing roles (`roles.yaml`) or default instructions only
|
||
|
||
## Compatibility Notes
|
||
- Loader merges by role ID; ensure IDs don’t collide with existing `roles.yaml` (IDs are based on filenames)
|
||
- `defaults.md` remains the global instruction source and is unchanged by this migration
|
||
|
||
## Evidence / References
|
||
- Loader & schema: `pkg/prompt/types.go`, `pkg/prompt/loader.go`
|
||
- Prompts directory & compose: `prompts/README.md`, `docker/docker-compose.prompts.dev.yml`
|
||
|