How to use the Source Intake Scanner agent skill (intake)

What is an agent skill and what does intake make the agent do?

An agent skill is a folder with a SKILL.md file that gives an AI agent specific instructions for one kind of task. The intake skill is named intake and its description is: inspect configured Sources and propose evidence-backed Backlog work.

With this skill, the agent reads the Product named in your message, lists its Sources, Projects and existing Tasks, and inspects evidence. It only proposes work when the evidence supports an actionable item. Reading or changing a file alone is not a reason for a Task.

• Every Task must belong to a Project. The agent reuses a suitable Project or creates one through propose_company_project.
• Each proposal includes a concise title, a brief, the evidence links, and the observed content hash for each link.
• Duplicate proposals are avoided. Proposals never copy whole documents into briefs.
• The agent never accepts or completes work. Proposals stay in Backlog.

How do you install the intake skill?

Download the ZIP file from this page and unzip it. For a personal setup, place the unzipped folder in ~/.claude/skills/. For a single project, place it in the project's .claude/skills/ folder.

You can also upload the skill in the Claude apps. Go to Settings, then Capabilities, then Skills, and upload it there.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside your project
• Claude apps: Settings → Capabilities → Skills

How do you trigger the intake skill with a natural request?

You do not need a special command. Name the Product and ask the agent to scan its Sources and propose Backlog work. The skill reads the Product identified in your message, so include the product name clearly.

The skill also covers the case where nothing new is found. If all evidence is already recorded, the agent returns a quiet successful check.

• "Scan the sources for the Atlas product and propose backlog tasks for real issues."
• "Run an intake check on the Atlas product and tell me if any evidence is new."
• "Review Atlas's configured sources and propose tasks only where the evidence supports them."

What does the agent need before it can run?

The skill depends on Agno's company Product, Source, Project and Task tools. It calls propose_company_project, read_product_source and intake_product_source. It also needs Product Sources that the agent can access. Without these tools the skill is not useful.

• Agno framework with company Product, Source, Project and Task tools
• Configured Product Sources accessible to the agent
• Runs scripts: no
• Uses network: yes

How does the agent read sources and avoid duplicates?

For each Source, the agent calls read_product_source without a path to get a paged index that contains no document content. It uses pattern and next_offset to narrow discovery, then passes a returned path to read one document. For large documents it uses next_content_offset to continue. If the full-document content hash changes between fragments, it rereads the document.

When it proposes work, it sends a stable proposal_key for the underlying issue. It reuses that key across scans, file edits and wording changes. The tool verifies links and hashes and deduplicates within the Product. The agent must not retry with a new key to bypass deduplication. A failed read produces no proposal.

• Use read_product_source without a path for a content-free index.
• Reuse the same proposal_key for the same issue.
• If evidence changes before Intake, read the Source again and reconsider.

What about safety, permissions and authorization?

The skill only proposes work. It forbids accepting or finishing tasks. If auto_accept metadata names an executor, the agent treats it as a suggested target, not as permission to accept work. The server checks current human authorization separately, and proposals remain Backlog until that check succeeds.

Because proposals depend on that server-side check, misconfigured authorization could let suggested executors act on work. Review authorization settings before any unattended use with broad permissions. The skill does not run scripts but does use the network.

• Existing and completed Tasks remain unchanged.
• The skill does not accept or complete work, so do not rely on it for automatic finishing.
• Review the human authorization configuration on your server before running unattended.