hope
Reviewed
at commit d377fdc
New version, not yet reviewed
Caution: Starts a separate `claude -p` with bypassed permissions and full tool access in your project after turns that edit tracked files, to judge them.
Thinking discipline for AI work. Clarify intent, shape approach, consult experts, guard unsupervised work, hand work back so you own it. The slop hook needs…
0 upvotes
Log in to upvote or save- Made by
- Saad Shahd
- Pinned commit
-
d377fdc1dbf317fbddd6c530e692e94d3c19e6fb - What we checked
-
Adds thinking skills and commands (intent, shape, consult, deepen, target, freeze) and a fixed rule in the session's first prompt telling Claude to run free, reversible steps itself instead of asking. After a turn that edits tracked files, a separate Claude run on your account reviews them against your CLAUDE.md and shows a 'slop' button; accepting a finding edits the file. Nothing leaves your machine and Claude.
Review details
Static review of the source at the pinned commit; not run. Delta review of the changes since 9f17963; the rest carries over from the review at that commit. Changes: hooks.json now loads hooks/index.tsx, which registers a new reach.ts and the existing slop.tsx. reach.ts now registers prompt.submit instead of prompt.compose: it reads the session messages ($.session.messages, as api) and, when none contains the fixed rule text yet (a first prompt, or after compaction), adds that constant text (turn to the user only for what nothing in reach settles, take free reversible steps yourself, leave to the user only payment, approval, sign-in or irreversible acts their words have not asked for) to the prompt context. This is prompt text only and does not change permissions. slop.tsx: the prompt.submit handler is now filtered to the fix and addressed-findings phrases, otherwise unchanged. findings from a judge run now merge into waiting ones (addFindings); accept writes the suggested change with $.fs.write only when the old lines are found in exactly one place, otherwise (file moved) the finding goes to Claude as one to fix instead of throwing; accepted-and-written changes are listed to Claude in the next submitted prompt when the box keeps the added 'Addressed slop findings' line, with a note to commit them; the pane shows a notice when the file has moved. Still registers turn.complete, prompt.submit, ui.render (AbovePrompt, Pane) and ui.close, reads session messages, and runs git status and hooks/judge.sh through $.process. judge.sh is unchanged in this diff and still runs `claude -p --permission-mode bypassPermissions` with hooks disabled and no tool allowlist; the prompt tells it to review and report only, and it is not otherwise fenced. Skill and command changes are prompt text only (new deepen skill; edits to clarify, consult, draft, elicit, explain, intent, inversion, reduce, reframe, router, shape, target); no allowed-tools, installs or URLs added. State is session-only. Setting CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1 is still required for the hooks. Version 13.9.2 also changes the shape skill (prompt text only: change no file while a choice is open, ask all open choices at once).
- What it uses
-
- Reads your prompts: yes
- Sees tool calls: yes
- Blocks or changes tool calls: no
- Reads or writes files: yes
- Runs programs on your machine: yes
- Calls the model (uses tokens): yes
- Sends data over the network: no
- Stores data: no
- Shows something on screen: yes
Thinking discipline for AI work. Clarify intent, shape approach, consult experts, guard unsupervised work, hand work back so you own it. The slop hook needs CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1.
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/hope/)