Agent-Skills.md

Agent Skills: Shepherd - Autonomous Pipeline Monitor

Autonomous pipeline monitor using sense-think-act loop. Watches snakemake/nextflow jobs, detects errors, applies fixes from memory, restarts on failure.

UncategorizedID: genomewalker/cc-soul/shepherd

Repository

genomewalker/cc-soul
genomewalker
33

Install this agent skill to your local

pnpm dlx add-skill https://github.com/genomewalker/cc-soul/tree/HEAD/skills/shepherd

Skill Files

Browse the full folder contents for shepherd.

Download Skill

Loading file tree…

skills/shepherd/SKILL.md

Skill Metadata

Name
shepherd
Description
Autonomous pipeline monitor using sense-think-act loop. Watches snakemake/nextflow jobs, detects errors, applies fixes from memory, restarts on failure.

Shepherd - Autonomous Pipeline Monitor

Tends long-running pipelines (snakemake, nextflow) using a sense-think-act loop. Detects errors, recalls fixes from memory, restarts automatically, checkpoints progress.

Quick Commands

/shepherd <command> [--interval=60] [--max-restarts=3]  # Start monitoring
/shepherd status                                         # Check shepherd status
/shepherd stop                                          # Stop monitoring

Configuration

| Flag | Default | Description | |------|---------|-------------| | --interval | 60 | Seconds between sense cycles | | --max-restarts | 3 | Max automatic restarts before escalating | | --notify | true | Send notifications on events | | --auto-fix | true | Attempt automatic fixes from memory | | --session | auto | Target session: "auto", "current", or session name | | --isolate | auto | Session isolation: "auto", "yes", "no" |

--isolate=auto picks isolation by job type (SLURM/remote-SSH/multiple parallel pipelines get an isolated session; quick local or interactive jobs don't). Read reference/configuration.md before implementing this decision — it has the full condition table and the shouldIsolate logic.

Core Workflow

  1. Initialize — load the zellij_mcp/chitta tools via ToolSearch, check for an existing long_task to resume, decide session isolation, create the monitoring pane, and launch the pipeline.
  2. Sense — tail the pane, check for idle/stall, search for error patterns.
  3. Think — match sensed output against known habits and pattern libraries; decide to apply a habit, restart, fix from memory, checkpoint, or escalate.
  4. Act — execute the decision: restart the command, apply a fix, log an event, notify, or mark the task complete/paused.
  5. Main loop — repeat sense/think/act on --interval, with a pre-flight health_check and session-liveness check each cycle, until done or paused.

Read reference/initialize-and-loop.md before implementing any of the steps above — it has the full pseudocode, including every mcp__zellij_mcp__* and mcp__chitta__* call.

Read reference/patterns-and-tools.md for known snakemake/nextflow error signatures, the expected status-line output format, and the full MCP tool list.

Read reference/resume-protocol.md when resuming a shepherd task after a disconnect or session restart.

Anti-Patterns

  • Never poll faster than 30 seconds (wastes resources)
  • Never restart without checkpointing first
  • Never exceed max_restarts without escalating
  • Never ignore repeated errors (they compound)
  • Never fix without logging (lose learnings)

Dashboard

/shepherd dashboard opens an interactive control center across all active shepherd tasks — view task details or pane output, restart/stop/escalate a pipeline, or view learned habits. Read reference/dashboard.md for the full implementation before building or running it.