capcom compared

capcom is one of several ways to track the work of AI coding agents. What sets it apart is that the files in git are the only store: no database to sync, no hosted service, no background process and no LLM calls of its own. This page compares it with the alternatives people most often weigh.

At a glance

	capcom	beads	GitHub Issues	Task Master	TODO.md or agent memory
Where the data lives	Markdown files in the repository	A database, with a JSONL export in git	GitHub’s servers	A JSON task file in the repository	One markdown file
Branches with the code	Yes	Through the export	No	Yes	Yes
Works offline, no account	Yes	Yes	No	Needs an LLM API key to generate tasks	Yes
Dependencies and a ready queue	Yes	Yes	Partial	Yes	No

Beyond the table, capcom validates its files in CI with capcom check, keeps project knowledge as small Notes beside the Issues, and shows every agent session who holds the Tree lock on a worktree.

capcom vs beads

beads (bd) is an issue tracker for coding agents that keeps its issues in a database and exports a JSONL copy for git. capcom has no database to keep in step: each Issue is its own markdown file, the SQLite index is a cache rebuilt from the files and never committed, and an Issue branches, merges and shows up in code review with the code it describes. Ids are readable slugs you choose (login-timeout-race) instead of hashes. To move a repository over, follow Migrating from beads.

capcom vs GitHub Issues

GitHub Issues lives on GitHub’s servers, so an agent needs network access and credentials to read it, and an issue cannot change on a branch and merge with the code. capcom lives in the repository: an agent reads it offline with no account, a pull request shows the Issue change next to the code change, and capcom ready answers “what is unblocked” from Dependencies. GitHub Issues remains the better place for bug reports from outside contributors; capcom is for the work the team and its agents plan and do.

capcom vs Task Master

Task Master is an AI task manager that keeps tasks in a JSON file and uses an LLM, with your API key, to break a product requirements document into tasks. capcom calls no LLM: the agent you already run writes Issues as markdown, and capcom validates, indexes and orders them. Each Issue is its own file, so two branches touching different Issues merge without conflict.

capcom vs a TODO.md or an agent instructions file

A single TODO.md, or a growing CLAUDE.md or AGENTS.md, works for a short list but has no Dependencies, no ready queue, no validation, and conflicts whenever two branches edit it. capcom splits work into one file per Issue and knowledge into small Notes an agent opens only when it needs them, and keeps AGENTS.md to a short block that points at them. See Knowledge and The Managed block.

When capcom is not the right fit

• You need a hosted board for people outside the repository, with notifications: capcom sends nothing and runs no server.
• Your issues must live apart from any one repository: capcom keeps each repository’s issues in that repository.