Configuration
The AgentLogs CLI stores configuration in~/.config/agentlogs/.
Settings File
Repository permissions are configured insettings.json:
The settings file uses JSONC format, which supports comments. Feel free to add notes explaining your configuration
choices.
Settings Fields
Sessions with Multiple Directories
For Claude Code and Codex JSONL uploads, capture permissions are checked across the session’s recorded working directories before upload:- In allowlist mode, every discovered Git root must have at least one allowlisted remote. A root with no remote, an unreadable config, or an unsupported remote URL blocks the whole upload.
- In denylist mode, any explicitly denied remote blocks the whole upload, including when another remote or directory is allowed.
- Directories outside Git repositories do not add another Git root to check. A session that starts in your home directory and then works in an allowed clone can upload. If no repository is found anywhere, allowlist mode skips it and denylist mode permits it.
origin is preferred when allowed; otherwise another allowed remote, such as upstream, is used. The selected repository also supplies the visibility setting and repository links.
For a fork with a personal origin and an allowlisted upstream, add the upstream’s normalized repository identifier to repos, for example github.com/myorg/project. Use agentlogs settings to edit these entries. agentlogs allow still applies to the current clone’s origin; Claude Code’s commit-link permission check also uses that origin.
Linked Git worktrees use their shared repository config and their own branch. Directory discovery reads Claude Code’s record-level cwd and Codex’s session_meta and turn_context payloads. Discovery and hook directory hints supplement these recorded directories for permission checks. AgentLogs does not infer directory changes from shell commands or message text.