Capability: dev-environment

How an agent session is set up in this repository: the instruction files it reads, the Claude Code settings it runs under, and the-loop's per-repo config.

What it is

The repository's developer environment for coding agents. Claude Code and other agent harnesses (via AGENTS.md) get the same instructions; Claude Code also installs the the-loop plugin from the repository's settings.

Current behaviour

• The repository SHALL carry AGENTS.md as the harness-neutral instructions, and CLAUDE.md SHALL import it (@AGENTS.md) and add only Claude Code specifics.
• .claude/settings.json SHALL match the-loop's own .claude/settings.json: the same permission allow-list, the the-loop marketplace (github: MadaraUchiha-314/the-loop) and the-loop@the-loop enabled.
• .the-loop/ SHALL hold exactly three files: harness-config.yaml, collaborators.yaml and manifest.yaml. The the-loop CLI is not configured here: no cli-config.yaml, no Slack manifest.
• .the-loop/collaborators.yaml SHALL name @MadaraUchiha-314 as the approver.
• The docs trees the-loop expects SHALL exist: docs/specs/, docs/capabilities/, docs/architecture/, docs/decisions/ and docs/learnings/.

Design

No design.md: issue-1 declared design away. The file layout follows the-loop's /the-loop:init command and its manifest.

History

Work item	What changed	Links
issue-1	Initial dev environment: agent instructions, Claude settings, the-loop init	spec, issue #1