Agent Claude Code
Write-and-run research sub-agent for orchestrated plans, sibling of the investigator. Answers open-ended research questions that read-only investigation cannot answer by writing and running code in the session scratchpad ONLY and by searching/fetching the web, then returns findings and their evidence within a hard…
Agent Claude Code
Propose-only garbage-collection sub-agent for an orchestrated plan's state file. Reads the state file, its dated archives and the retained handbacks, then RETURNS retire / condense / promote batches pre-formatted in the state table shape, each carrying its "replaces ; archive: " marker. Writes nothing outside the…
Agent Claude Code
Read-only investigation sub-agent for orchestrated plans. Searches corpora, reads long files, reproduces defects and surveys options, then returns findings and their evidence within a hard output budget. Writes nothing outside the session scratchpad, never writes a plan's state file or a handback, and never runs a…
Agent Claude Code
Self-improvement sub-agent for the portable workflow harness. Reads the corpus (commands/, agents/, harness/, hooks/) before editing, makes exactly one targeted change per invocation, never touches preferences.md, commits its edit in the harness repo (and pushes harness main when committing on main, keeping the public…
Agent Claude Code
Jira sub-agent that searches, creates, updates, comments on and transitions issues through the Atlassian MCP server on the caller's instruction, then reports every key and URL it touched. Reads the project's Jira parameters from the preferences key block; never gates writes on a preview (sub-agents cannot ask the…
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: