alex/g3
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: Externalize research tool as embedded skill

Browse Source 
Replaces the built-in research/research_status tools with a portable
skill-based approach:

- Add embedded skills infrastructure (skills compiled into binary)
- Add repo-local skills/ directory support (highest priority)
- Create research skill with SKILL.md and g3-research shell script
- Script extraction to .g3/bin/ with version tracking
- Filesystem-based handoff via .g3/research/<id>/status.json
- Remove PendingResearchManager and all research tool code
- Update system prompt to reference skill instead of tool

Benefits:
- No special tool infrastructure needed (just shell + read_file)
- Context-efficient (reports stay on disk until needed)
- Crash-resilient (state persisted to filesystem)
- Portable (skill can be overridden per-workspace)

Breaking change: research tool calls now return a deprecation message
pointing to the research skill.
This commit is contained in:
Dhanji R. Prasanna
2026-02-05 13:23:26 +11:00
parent bf9e3dc878
commit 39e586982c
19 changed files with 949 additions and 1638 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
prompts/system/native.md
View File

@@ -160,20 +160,24 @@ facts:
# Web Research
When you need to look up documentation, search for resources, find data online, or research a topic to complete your task, use the `research` tool. **Research is asynchronous** - it runs in the background while you continue working.
When you need to look up documentation, search for resources, find data online, or research a topic to complete your task, use the **research skill**. Research runs in the background while you continue working.
**Use the `research` tool** for any web research tasks:
**Use the research skill** for any web research tasks:
- Researching APIs, SDKs, libraries, frameworks, or tools
- Finding approaches, patterns, or best practices
- Investigating bugs, issues, or error messages
- Looking up documentation or specifications
**How async research works:**
1. Call `research` with your query - it returns immediately with a `research_id`
**How to use the research skill:**
1. Start research with `background_process`: `background_process("research-topic", ".g3/bin/g3-research 'your query'")`
2. Continue with other work while research runs in the background (30-120 seconds)
3. Results are automatically injected into the conversation when ready
4. Use `research_status` to check progress if needed
5. If you need results before continuing, say so and yield the turn to the user
3. Check status: `shell(".g3/bin/g3-research --list")` or `shell(".g3/bin/g3-research --status <id>")`
4. Read the report when ready: `read_file(".g3/research/<id>/report.md")`
**Key points:**
- Always use `background_process`, never blocking `shell` for research
- Reports are saved to disk, not auto-injected - read them when you need them
- Check `.g3/research/<id>/status.json` for machine-readable status
# Workspace Memory