with one click
manage-skills
// Autonomously create and update persistent skills in the /workspace/skills/ directory
// Autonomously create and update persistent skills in the /workspace/skills/ directory
Using the /events/ directory to schedule one-shot (at) or periodic (cron) jobs that fire asynchronously, possibly in a thread
Reconstruct high-level past events (decisions, commitments, timelines) from /chronicle chapters. Use for longer-term recall beyond what <context_summary> paragraphs cover — arc of past discussions, decisions made, what happened last week(s).
Using /workspace/HEARTBEAT.md for periodic self-directed background work and proactive checks
| name | manage-skills |
| description | Autonomously create and update persistent skills in the /workspace/skills/ directory |
Skills persist across sessions. You are in charge of maintaining them, and you can continuously learn by creating new skills (or updating existing ones with new lessons!). The caveat is that having too many skills will waste your context.
You can create / edit / etc. the files /workspace/skills/<name>/SKILL.md in
agentskills.io format with YAML frontmatter:
---
name: <name>
description: <one-line>
---
<procedural instructions>
name must match directory name (lowercase, a-z, 0-9, hyphens).description is the lead inserted in every system prompt - must be 1-1024 characters and describe both what the skill does and when to use it, should include specific keywords that help agents identify relevant tasksBody content should include step-by-step instructions to perform the skill, examples of inputs and outputs, common edge cases. But be very terse, these are notes you jot down for yourself, not ELI5 teaching material - focus just on key directions and lessons hard learned. Never more than ~100 lines.
But additional resources that you can reference or execute as needed can live
in scripts/, references/ and assets/ in your skill directory alongside
SKILL.md. Reference them with relative paths in the skill body.