Skill Lifecycle
Use this skill to identify, author, register, and maintain reusable skills across the organization. It is the process behind all other skills.
When to Use
- After completing any Paperclip task (post-task skill check)
- During quarterly skill audits
- When asked to create or update a skill
- When onboarding a new agent that needs skills assigned
Step 1 — Identify Skill Candidates
After every completed task, run the 3-question check:
- Was this process repeatable?
- Did it involve 3+ discrete steps?
- Would another agent need to do this again?
If yes to all three, file a skill-creation subtask under the parent goal.
Where to Look
- Completed Paperclip tasks (especially multi-step ones)
- Recurring manual processes (triage, deployment, review)
- Patterns that multiple agents follow independently
- Ad hoc audits of recent work
Step 2 — Author the Skill
Canonical Location
All skills live at:
/workspace/.ai/skills/{skill-name}/SKILL.md
Optional supporting files go in a references/ subfolder.
SKILL.md Structure
Use this template:
---
name: {skill-name}
description: >
One-line description of what the skill does and when to use it.
This is used for relevance matching, so be specific.
---
# {Skill Title}
{Brief introduction — what this skill does and why it exists.}
## When to Use
{Bullet list of trigger conditions.}
## Prerequisites
{Environment variables, tools, or permissions required.}
## Workflow
{Numbered steps with code blocks, API calls, or procedures.
Each step should be concrete and actionable.}
## Notes
{Edge cases, caveats, or integration points.}Authoring Guidelines
- Be concrete. Include actual commands, API calls, and templates — not abstract advice.
- Be self-contained. A skill should work without requiring the reader to
search for external docs. Use
references/for large supporting files. - Extract from real work. Skills come from completed tasks, not invention. Reference the source task in the description or notes.
- One canonical source. Other tool integrations (Claude, Paperclip, etc.) link to the SKILL.md rather than duplicating content.
Step 3 — Register in Paperclip
Option A: Scan Projects (preferred)
If the skill is in a project workspace that Paperclip knows about:
POST /api/companies/{companyId}/skills/scan-projectsThis discovers all SKILL.md files in known project workspaces.
Option B: Manual Import
POST /api/companies/{companyId}/skills/import
{
"skills": [
{
"name": "{skill-name}",
"description": "{description}",
"path": ".ai/skills/{skill-name}/SKILL.md"
}
]
}Refer to skills/paperclip/references/company-skills.md for full import API details.
Step 4 — Assign to Agents
After registration, assign the skill to relevant agents:
POST /api/agents/{agentId}/skills/sync
{
"desiredSkills": ["companyId/{skill-name}", ...]
}Choose agents based on who performs the work the skill encodes:
- CEO: skill-lifecycle (for post-task checks), prd-authoring
- CTO: all technical skills
- Engineers: skills relevant to their domain (CI, repo setup, etc.)
- PO/Triage agents: triage-related skills
Step 5 — Maintain
- Update skills when the underlying process changes (new tools, new APIs, new team structure).
- Retire skills that are no longer needed — delete the SKILL.md and remove from agent assignments.
- Version significant changes in a commit message referencing the skill name.
- Quarterly audit: review all skills for accuracy and completeness. Check completed work for missed skill candidates.
Integration with CEO Heartbeat
The CEO agent should run the 3-question skill check as part of every task closure. When a candidate is identified:
- Create a subtask under the relevant goal with title:
Create {skill-name} skill - Assign to the CTO or appropriate engineer
- Set priority based on frequency and impact
Notes
- Skills are tool-agnostic — they encode process, not tool configuration
- The canonical path
/workspace/.ai/skills/is shared across all tools - Registration in Paperclip makes skills discoverable and assignable
- This skill is itself an example of the pattern it describes