How to use the Live Research Project Manager agent skill

What the skill makes the agent do

The skill turns the agent into a post-task research recorder. It runs only at the end of a session, after your request has been fully addressed. It reviews what happened in the conversation and updates an ara/ directory so the project's history is written down.

It looks for specific event types: decisions (a choice between alternatives), experiments (a test or benchmark with a result), dead ends (an approach that was abandoned), pivots (a major change of direction based on evidence), claims (assertions about the system), heuristics (implementation tricks), AI actions (code written or commands run) and unclassified observations. Routine file reads, typo fixes, git operations and dependency installs are skipped.

• Decisions, experiments, dead ends and pivots go to trace/exploration_tree.yaml.
• Claims go to logic/claims.md.
• Heuristics go to logic/solution/heuristics.md.
• Unclassified observations go to staging/observations.yaml.
• AI actions are recorded in the session record only.

Understand the provenance tags

Every entry carries a provenance tag so the record shows who decided what. The source lists four tags: user (you stated or confirmed it), ai-suggested (the AI inferred it and you did not confirm), ai-executed (the AI performed the action) and user-revised (the AI suggested something and you corrected it).

When the agent is unsure, it defaults to ai-suggested and is not supposed to mark an inference as user. Treat the tags as a statement of origin, not as proof that a result is true.

Install the skill

You can install the skill in one of three ways. Download the ZIP from the skill's page on this site and unzip it into your user skills folder, ~/.claude/skills/, so it is available across projects. Or unzip it into a project's .claude/skills/ folder so it applies only to that project. Alternatively, upload the ZIP in the Claude apps under Settings → Capabilities → Skills.

After installing, the folder should contain SKILL.md along with the references/ folder (event-taxonomy.md, provenance-tags.md and session-protocol.md).

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

Check what it needs

The skill needs an agent that can read and write files in the project directory. The source lists no external libraries, API keys or accounts. It does not run scripts and does not use the network, according to the source.

• Read and write access to the project directory
• No external libraries, API keys or accounts listed
• Runs scripts: no
• Uses network: no

Trigger the skill with natural requests

The skill is meant to run at the end of a session, not during the work. Once your task is complete, ask the agent to record the session. The skill's description says to use it as a session epilogue and never during execution, so ask for it only after the work is done.

The source does not list a fixed set of trigger phrases, so the wording below is a suggestion of the kind of request that fits the purpose described in the source.

• "We're done with this task. Please record the research provenance for this session in ara/."
• "Wrap up: capture the decisions, dead ends and results from today's session."
• "Update the ara/ log with what we tried and why we changed direction."

Read and check the output

When the epilogue finishes, the agent reports a one-line summary of what it captured. Open the ara/ files and check that each entry has the correct tag. Correct any entry the skill marked wrongly, since the record is only as faithful as what the conversation contained.

The skill records what the conversation shows and labels inferences as AI-suggested. It does not verify that recorded results are true, so check any numbers against your own evidence.

• Confirm user-stated items carry the user tag.
• Look at ai-suggested items and remove any you do not agree with.
• Check recorded numbers against your own runs.