How to use the Structured Note-Taking agent skill (hive.note-taking)

What the note-taking skill makes the agent do

The skill instructs the agent to maintain free-form working notes in a shared buffer key named _working_notes. These notes hold data the agent needs to remember but that the colony task queue does not capture.

The skill updates the notes at three checkpoints: after new information changes the plan for the current step, before a tool call that will produce substantial output the agent will need later, and when the agent makes a non-obvious decision whose reasoning would be lost if tool-call history is pruned. Updates are incremental, not a full rewrite each time.

• Key Decisions: decisions made and why
• Working Data: intermediate results and extracted values such as URLs, IDs and key snippets, not full pages
• Open Questions: uncertainties the agent plans to verify
• Blockers: anything preventing progress that is not already recorded in tasks.last_error

What the skill does not do

The skill is for the agent's internal bookkeeping. It does not replace the task queue. Per-task goals, ordered steps and SOP gates belong in progress.db, which the skill says to manage with hive.colony-progress-tracker.

The notes live in the Hive agent's shared buffer. They are not a personal note app and are not visible as a document you can browse outside that buffer.

Install the skill from a ZIP file

Download the skill ZIP from this page and unzip it. The package contains a single file, SKILL.md. Place the unzipped skill folder in one of these locations, depending on how you want to use it:

For a user-wide install, put it in ~/.claude/skills/. For a single project, put it in the project's .claude/skills/ folder.

In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

• User-wide: ~/.claude/skills/
• Single project: .claude/skills/ inside the project folder
• Claude apps: Settings → Capabilities → Skills, then upload

Check the requirements before installing

This skill only makes sense inside the Hive agent framework. The source says it needs a Hive agent framework with a shared buffer and a progress.db / tasks tracker, and it refers to hive.colony-progress-tracker. Without those pieces the _working_notes key and the checkpoints have nothing to work against.

The skill has no scripts and makes no network calls, according to the source.

• Hive agent framework with a shared buffer
• progress.db and a tasks tracker
• hive.colony-progress-tracker skill for task-level state
• Runs scripts: no
• Uses network: no

Trigger the skill with natural requests

The skill is marked for the worker role and is applied as operational guidance while the agent works. You do not need a special command. In a Hive worker, the agent follows the protocol during multi-step runs, especially when it makes many tool calls or is near context limits.

The skill's own description is the best guide to what triggers it: it applies when the agent needs to keep decisions, extracted values and open questions so context pruning does not lose them. You can also ask for it directly in a prompt, for example: "Track the IDs you extract and record why you chose each option in your working notes."

• Keep track of every URL and ID you extract, and note why you picked each one.
• Record your open questions so you can verify them before you finish.
• Note any blocker in your working notes, and check the task queue for errors separately.

Example: a long-running research agent

Imagine a Hive worker that searches sources, fetches several pages and calls many tools before writing a report. Early on it chooses one data source over another. With the skill active, it records that decision and the reason under Key Decisions, saves the URLs it will cite under Working Data, and lists a figure it has not yet confirmed under Open Questions.

Later, when older tool results have been pruned, the agent can read _working_notes and still know why it chose that source and which values to check.

Safety notes: keep secrets out of the notes

The source gives no guidance on excluding secrets or personal data. Anything written to _working_notes sits in a shared buffer that other parts of the system may read. Do not store passwords, API keys, tokens or personal data there.

Review what your agent writes to the buffer, especially if other agents or tools can read it.

• Do not write credentials or personal data into _working_notes.
• Assume anything in the shared buffer can be read by components that use it.

Tips for getting good notes

Keep entries short. The skill asks for extracted values, not full pages, so store a URL or ID rather than the page contents.

Use the notes only for information the task queue does not already hold. Duplicating goals and steps in the notes adds overhead without benefit.

Skip the skill for simple one-shot chats where context pruning never happens, because the notes then add cost with no gain.

• Store the value and the reason, not the whole document.
• Do not copy task steps from progress.db into the notes.
• Use it for long multi-step runs, not short exchanges.