How to use the Context Preservation agent skill (hive.context-preservation)

What does the Context Preservation skill make the agent do?

An agent works within a finite context window. When that window fills up, older tool results are pruned, which means the agent can no longer see them. The skill tells the agent to extract the information it will need later while the result is still in context.

Right after a tool call that returns information the agent will need, it writes the key data into _working_notes or _preserved_data instead of relying on re-reading the old output.

• Keep URLs and short key snippets, not full pages.
• Keep relevant API fields, not raw JSON blobs.
• Keep specific lines, values or IDs, not entire files.
• Keep analysis conclusions, not raw data.
• Pass state between tasks through progress.db rather than shared-buffer handoff blobs.

How do you install the Context Preservation skill?

The skill is distributed as a ZIP file from the directory page. It contains a single file, SKILL.md. You can install it in one of two ways.

For Claude Code or a similar setup, unzip the download into ~/.claude/skills/ to make it available everywhere, or into a project's .claude/skills/ folder to limit it to that project.

In the Claude apps, upload the ZIP under Settings → Capabilities → Skills.

• Personal: unzip into ~/.claude/skills/
• Project only: unzip into the project's .claude/skills/ folder
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

How do you trigger the skill with natural requests?

The source does not list specific trigger phrases. The skill's description is about extracting critical values from tool results before they are pruned, so it is most relevant during long tasks that gather information across many tool calls, such as research or data collection.

You do not need special syntax. Describe the long-running task you want done and make clear that the agent must keep the key facts it finds.

• Ask for a multi-step research task and say the agent should keep track of sources and values it finds along the way.
• Ask for a data-gathering run where IDs and specific values from each step must be retained for later steps.
• Note that the skill is written for workers in the Hive framework, so it works best when the agent runs there.

What does the skill need to run?

The skill requires the Hive agent framework, including its progress.db task database and its note storage (_working_notes and _preserved_data). Outside Hive, the instructions refer to names that other setups will not have.

The skill does not run scripts and does not use the network, according to the item's listing. Its only file is SKILL.md.

• Hive agent framework with progress.db and note storage
• No scripts are run by the skill
• No network use

What are the safety notes?

Because the skill runs no scripts and uses no network access, its risk is low. It is a text instruction file that changes how the agent records information.

Be aware that the agent will write facts into notes and task records. If those records contain sensitive values, review what is stored in progress.db and the note fields.

• No scripts are executed by this skill.
• No network access is used by this skill.
• Review stored notes and task rows if they may contain sensitive data.

Tips for getting good results

The skill helps most when context is likely to fill up. For a short conversation with a single step, the extra note-taking adds little value.

When the agent finishes a task, the next worker reads state from the task row (steps.evidence, tasks.last_error, sop_checklist.note), and _working_notes covers anything the database schema does not. The source also says the agent receives an alert when context reaches a set usage ratio, at which point it should preserve data immediately.

• Use it for long, multi-step work, not one-off questions.
• Expect state handoffs between tasks to go through progress.db.
• Use _working_notes for information the database schema does not cover.