How to use the Knowledge Ops SOP Toolkit agent skill

What is an agent skill and what does knowledge-ops make the agent do?

An agent skill is a folder with a SKILL.md file and supporting files that an AI agent loads when a request matches the skill's description. knowledge-ops is aimed at Heads of Operations, Knowledge Managers and TPM-Internal roles who need to author, validate or clean up company procedures.

Once installed, the agent can write SOPs and runbooks in a 5W2H structure (Who, What, When, Where, Why, How, How-much), check existing runbooks step by step, and scan an exported wiki (Notion, Confluence, Obsidian or a folder of markdown) for orphan pages, stale pages and inconsistent glossary terms. It then produces a ranked list of what to fix first.

• Author SOPs for processes such as procurement intake, vendor offboarding, incident-comms cascade and employee onboarding.
• Validate runbook steps for a named owner, expected duration, observable success and failure signals, rollback path and escalation contact.
• Ingest a wiki export and report orphan pages, stale pages, glossary drift and missing owners.
• Get a prioritized cleanup list of the documents to fix first.

How do you install the knowledge-ops skill?

The skill is distributed as a ZIP file from the directory page. Install it in one of two ways. For Claude Code, unzip it into your personal skills folder ~/.claude/skills/ or into a project's .claude/skills/ folder so it applies only to that project. In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.

The skill bundles Python scripts, so you also need Python 3 available on the machine where the agent runs. The scripts use only the standard library, so there are no packages to install.

• Claude Code, personal: unzip into ~/.claude/skills/
• Claude Code, single project: unzip into the project's .claude/skills/
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

How do you trigger the skill with natural requests?

You do not need a special command. The skill's description covers requests about SOPs, runbooks, 5W2H checks, wiki audits and runbook verification, so describe the task the way you would to a colleague. Give the agent the file or export it should work on, because the scripts need a markdown or JSON input.

Good requests name the document and the goal. For example, ask the agent to validate a specific runbook before it goes into rotation, or to audit a wiki export for stale and orphaned SOPs.

• "Validate this incident-comms runbook before it goes into rotation."
• "Audit our Confluence wiki export for stale and orphaned SOPs and give me the top 20 to fix."
• "Draft a vendor offboarding SOP in 5W2H form for a Vendor Management operator."
• "Check every step in this runbook for a named owner and a rollback path."

What do the bundled scripts check?

The skill includes three scripts. sop_generator.py reads a JSON metadata file describing an SOP and emits a 5W2H-structured SOP in markdown or normalized JSON; its --profile option accepts ops, support, finance, hr, it or regulated. runbook_validator.py checks each runbook step against six attributes and gives a per-step GREEN, AMBER or RED rating, an overall score from 0 to 100 and a MUST-FIX list. kb_ingester.py walks a folder of markdown files and reports cross-links, orphan pages, glossary drift, stale pages (no edit in more than 12 months, from file modification time or a last_reviewed field), missing owners and a top-20 cleanup list.

The validator verdict bands are 80 or above for SAFE-TO-USE, 60 to 79 for USE-WITH-CAUTION, and below 60 for NOT-SAFE.

• Named owner, not 'the team' or 'ops'
• Expected duration as a number with a unit
• Observable success signal and observable failure signal
• Rollback path, or an explicit statement that the step cannot be rolled back
• Escalation contact, named person or named on-call rotation

What does the skill need and what are the safety notes?

The skill needs Python 3 and a markdown or JSON export of the wiki or runbook you want to analyse. It runs scripts, and it does not use the network according to its listing. Review the scripts before you run them on sensitive content, as you would with any script an agent executes.

The validator checks structure, not technical correctness. A runbook can score well and still contain a wrong command, so an engineer should confirm the procedure itself. The regulatory sections are templates and scaffolds only. They reference ISO 9001, SOC 2, HIPAA and FDA but do not certify compliance.

• Requires Python 3; scripts use only the standard library.
• Runs scripts; uses no network according to the listing.
• A high score is not proof a live runbook is safe.
• Regulatory output is a scaffold, not an audit or certification.

What is a practical workflow for a wiki cleanup sprint?

Start by ingesting the wiki export so you know which documents are unfindable or stale. Then validate the runbooks on the cleanup list, generate new SOPs where existing ones cannot be salvaged, and re-run the ingester after the sprint. The skill recommends tracking orphan-page count and the number of unsafe runbooks rather than total page count.

• Ingest the export with kb_ingester.py to get the health report and top-20 list.
• Validate each runbook with runbook_validator.py.
• Generate replacement SOPs with sop_generator.py where needed.
• Re-run kb_ingester.py to confirm orphans are down and glossary drift is resolved.