inbox
Reviewed
at commit dc47710
Collects what is waiting on you in a session: questions, Claude's notes and PR reviews
0 upvotes
Log in to upvote or save- Source
- github.com/petekp/inbox
- Made by
- petekp
- Pinned commit
-
dc477109fda29e3b2ef03ea30af9c0a6dad266f2 - What we checked
-
After each of Claude's replies, it asks Sonnet on your own Claude Code account to update a list of open questions, tasks, findings and failed tests, shown in a line above the prompt and in an /inbox panel. It looks up your pull requests through the GitHub CLI and makes Claude run tests or builds as separate commands, keeping their output in log files.
Review details
Static review of the source at the pinned commit; not run. Registers session.start, session.attach, classic.SessionStart, session.end, prompt.submit, turn.start, prompt.compose (adds a guidance section to the system prompt), tool.describe, tool.check, tool.call, classic.PermissionRequest (only shows the pending prompt in its band, never answers it), classic.StopFailure, classic.Stop, turn.complete, prompt.context (adds the current inbox text to each request), session.append, command.run (/inbox and /inbox demo), ui.close, ui.render (AbovePrompt band and a Pane) and config.set (theme). It registers three tools of its own (mcp__inbox__record_finding, close, run_check) and auto-allows them; run_check runs each check through $.tool.call Bash, so the normal Bash permission check still applies, and Read and Grep are auto-allowed only on its own check-log folder. It denies Claude's own Bash calls that combine a recognised test, type check, lint or build command with a pipe, redirect or other command, asking for it to be run alone, and classic.Stop sends Claude back once when the reply claims a check passes that last failed or has not run since an edit. Model use: $.model.complete with sonnet after each answered turn (prompt words, tool activity lines, reply, check results) and $.model.fork or $.model.complete over $.session.messages for catch-ups, all on the user's own account. Programs: gh pr view and gh api graphql (read-only query for review threads) in the session folder; read-only git commands (rev-parse, status, ls-tree, diff, hash-object without -w); macOS open for a URL (https only, validated) or a path (executables and app bundles are only revealed in Finder) on a button press; test -x; and herdr pane report-metadata only when HERDR_PANE_ID is set (a terminal multiplexer integration not mentioned in the README). Writes check output logs to the repo's .git/inbox/checks/<session> or $TMPDIR/inbox-checks/<session>. Stores the ledger per session (last 40 kept) and per project folder in $.store; reads the theme via $.config.list and HOME, TMPDIR and HERDR_PANE_ID. Buttons send prompts as the user's message (answers, fix this finding, address review comments without posting, draft a reply without posting), so Claude's usual permission prompts apply to what follows. No fetch or other network code. The truncated register.tsx was read from the clone; register.tsx sections with hooks, process, file, model and permission code, presses.ts, prs.ts prompts and queries, tools.ts, tree.ts and the check patterns in checks.ts were read fully; the remaining pane rendering (register.tsx lines 2300-3735), ledger.ts, check-tracking.ts and demo.ts were covered by targeted search. The codex/ folder and scripts/check.sh are not loaded by the Claude Code plugin and were not reviewed.
- What it uses
-
- Reads your prompts: yes
- Sees tool calls: yes
- Blocks or changes tool calls: yes
- Reads or writes files: yes
- Runs programs on your machine: yes
- Calls the model (uses tokens): yes
- Sends data over the network: yes
- Stores data: yes
- Shows something on screen: yes
- Hooked events
Made this mod? Add a badge to your README
Paste this into your README. The badge says reviewed once the current version has been reviewed, and links readers to this page.
[](https://modsforclaude.com/mods/inbox/)