Remediation

Skills

Reusable instructions for recurring investigations. Create a skill once and every thread and automation in the workspace can follow it.

A skill is a named set of instructions your agents load on demand. Each skill has a name, a slug, a description, an instructions file (shown as SKILL.md), and optional attached markdown documents such as runbooks and reference sheets. If you keep walking an agent through the same procedure, encode it once as a skill.

Create a skill

Open Settings, then Skills

Click Add skill.

Start from the catalog or from scratch

The catalog holds pre-built skills such as investigate-latency, incident-investigation, review-pr, daily-health-report, and security-assessment. Each one ships instructions plus reference documents, and installing it copies it into your workspace where you can edit everything. Pick Add your own instead to write a custom skill: name, slug, description, and instructions. The slug is generated from the name; agents reference the skill by this slug.

Attach documents

From the skill's page, add markdown documents. Agents see a manifest of filenames and descriptions and read only the documents they need, so a skill can carry a long runbook without bloating every conversation.

Worked example: a deploy-rollback runbook

Create a skill named Deploy Rollback with slug deploy-rollback:

Instructions
You are investigating a bad deploy and deciding whether to roll back.

1. Confirm the suspect deploy: query the deployment timeline for the
   affected service over the past 6 hours.
2. Chart error rate and 99th percentile latency before and after
   the deploy.
3. Read rollback-procedure.md for the exact rollback steps and their
   preconditions.
4. If the deploy caused the regression, propose the rollback and wait
   for explicit confirmation. Never roll back without a go-ahead.
5. After rollback, verify error rate returns to baseline and post a
   summary with the evidence.

Attach rollback-procedure.md with your team's actual steps:

rollback-procedure.md
# Rollback procedure: payments-api

Preconditions: no in-flight database migration from the suspect deploy.

1. Identify the previous healthy release tag.
2. Redeploy that tag through the normal pipeline.
3. Watch error rate for 15 minutes before declaring recovery.

In any thread, type /deploy-rollback and the agent follows the runbook.

When agents load skills

  • Every thread turn includes a briefing that lists the workspace's enabled skills, so the agent knows what is available.
  • Typing / in the thread composer opens the skill picker. Selecting a skill inserts its slug, and the agent loads that skill's instructions and document manifest.
  • Automations can attach skills. Each run's thread receives the skill instructions before the agent starts, and automation templates can name skills to load. The automation builder installs any referenced catalog skill that isn't in your workspace yet.
  • Polylane's own workflows load system skills, for example polylane-autofix during autofix runs.

Disabling a skill removes it from the briefing and the picker, and agents can no longer load it.

Share a skill publicly

Open a skill and choose Share. Skills default to workspace visibility; setting visibility to Public produces a link anyone can open:

https://console.polylane.com/{workspace_slug}/settings/skills/p/{skill_id}

The public page is a read-only view of the skill's SKILL.md and attached documents. Don't share personal information or third-party content without permission.

Try it from your coding agent:

Use Polylane to create a skill from my rollback runbook: I will describe the steps, and you turn them into a named skill with clear instructions that any thread can load.