Home/AI & LLMs/research-tracker

research-tracker

Safe
AI & LLMs

Manage autonomous AI research agents with SQLite tracking.

SKILL.md

# Research Tracker CLI tool for managing autonomous research agents with append-only state, instruction queues, and oversight. ## Prerequisites ```bash brew tap 1645labs/tap brew install julians-research-tracker ``` Or: `go install github.com/1645labs/julians-research-tracker/cmd/research@latest` ## Quick Start ### Start a research project ```bash research init market-q1 --name "Q1 Market Analysis" --objective "Analyze competitor pricing and positioning" ``` ### As the research agent — log progress ```bash export RESEARCH_SESSION_ID="$SESSION_KEY" # Track which agent is writing research log market-q1 STEP_BEGIN --step 1 --payload '{"task":"gather sources"}' # ... do work ... research log market-q1 STEP_COMPLETE --step 1 research heartbeat market-q1 ``` ### Check status (from main session or heartbeat) ```bash research status market-q1 --json research context market-q1 --last 5 # Truncated context for prompts ``` ### Send instructions to running agent ```bash research instruct market-q1 "Focus on enterprise segment" --priority URGENT research stop-signal market-q1 # Request graceful stop ``` ### Agent checks for instructions ```bash research pending market-q1 --json research ack market-q1 --all # Acknowledge after processing research check-stop market-q1 # Exit 0 = stop, Exit 1 = continue ``` ## Commands Reference | Command | Purpose | |---------|---------| | `init <id> -o "..."` | Create project with objective | | `list [--status active\|done\|all]` | List projects (includes `needs_attention` flag) | | `show <id>` | Project details + recent events | | `stop <id>` | Stop project, send STOP instruction | | `archive <id>` | Archive completed project | | `log <id> <event> [--step N]` | Log event (STEP_BEGIN, CHECKPOINT, BLOCKED, etc.) | | `heartbeat <id>` | Update alive timestamp | | `block <id> --reason "..."` | Mark blocked, needs input | | `complete <id>` | Mark done | | `status <id> [--json]` | Current state summary | | `context <id> [--last N]` | Truncated context for agent prompts | | `instruct <id> "text"` | Send instruction | | `pending <id>` | List unacked instructions | | `ack <id> [--all]` | Acknowledge instructions | | `check-stop <id>` | Exit code: 0=stop, 1=continue | | `audit <id> --verdict pass\|drift` | Log audit result | ## Event Types `STARTED`, `STEP_BEGIN`, `STEP_COMPLETE`, `CHECKPOINT`, `BLOCKED`, `UNBLOCKED`, `AUDIT_PASS`, `AUDIT_DRIFT`, `HEARTBEAT`, `DONE`, `STOPPED`, `TIMEOUT` ## Integration Pattern ### Spawning a research agent ``` 1. research init <project> --objective "..." 2. sessions_spawn with task including: - Project ID and objective - Instructions to use research CLI for state - Check stop signal before each step - Log progress with heartbeat 3. Heartbeat monitors: research list --json | check needs_attention 4. Send instructions via: research instruct <project> "..." ``` ### Agent loop (in spawned agent) ```bash while research check-stop $PROJECT; [ $? -eq 1 ]; do research pending $PROJECT --json # Check instructions research log $PROJECT STEP_BEGIN --step $STEP # ... do work ... research log $PROJECT STEP_COMPLETE --step $STEP research heartbeat $PROJECT STEP=$((STEP + 1)) done research complete $PROJECT ``` ## Attention Detection `research list --json` includes `needs_attention: true` when: - Last event is BLOCKED - Has unacked URGENT or STOP instructions - Heartbeat stale (>5 min since last HEARTBEAT event) - Last audit was AUDIT_DRIFT ## Database SQLite at `~/.config/research-tracker/research.db` (WAL mode, append-only events). Run `research db migrate` after install. Schema auto-migrates on first use.

More in AI & LLMs