Repository navigation
Add muse (Meta Muse Code) agent preset - #76
Merged
Merged
Conversation
Closes the muse row of rxdt#49: one agent per PR. - [tool.harness.agents] gains a muse preset. `muse exec` takes the prompt positionally (no native stdin), so ralph.sh's piped prompt is forwarded via the sh -c '"$(cat)"' wrapper. - --disable-approval silences approval prompts but keeps the OS sandbox on; --yolo is deliberately not used (it would drop the sandbox too). - Verified against the real CLI (Muse Code 1.4.0, installed from the official installer): flags confirmed via `muse exec --help`. - Auth is the user's own: `muse login` or META_API_KEY. - README: muse added to the agent lists plus a short section on how to edit/add an agent preset. - Test: agent set updated in test_every_supported_agent_has_a_nonempty_command. Local verification: piped a prompt through the exact preset argv as ralph.sh does; the prompt arrived intact (turn_input_user event) and a full JSONL stream completed with exit 0. Live model runs need the user's own Meta credentials; without them the CLI fails fast with a clear 'missing meta credentials' message.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Taking the muse (Meta) row of #49 — one agent per PR, as suggested.
What changed
pyproject.toml: newmusepreset in[tool.harness.agents].muse exectakes the prompt as a positional arg (no native stdin), so the prompt ralph.sh pipes to stdin is forwarded via thesh -c 'muse exec --json --disable-approval "$(cat)"'wrapper.--disable-approvalsilences tool-approval prompts but keeps the OS sandbox on —--yolowas deliberately not used (it drops approval and the sandbox, per your note in Add more agents to[tool.harness.agents]#49).harness/tests/test_cli.py: updatedtest_every_supported_agent_has_a_nonempty_commandto expect the new agent set.README.md: muse added to the agent lists/examples, plus the short "Adding (or editing) an agent preset" section the issue asked for.Install / docs
curl -fsSL https://dev.meta.ai/install.sh | bashmuse loginorMETA_API_KEY. Verified the CLI version I tested against is Muse Code 1.4.0.Local verification (the part most contributors can't do)
muse exec --help(--json,--disable-approval, positional prompt) — your sketch in Add more agents to[tool.harness.agents]#49 was accurate.sh -c 'muse exec --json --disable-approval "$(cat)"'): the prompt arrived intact in theturn_input_userJSONL event ("prompt":"PROOF-PING-123\n\nRALPH_ITERATION=1/1") and the run completed with exit 0 — 30 JSONL events streamed./tmp-disk-full template test and two vale tests — vale isn't installed in this sandbox). Ruff lint + format pass on the touched files.One honest caveat: I verified the full pipeline with Muse's offline echo provider, since live model runs need the user's own Meta credentials — which I obviously can't supply here. The good news for users: without credentials the CLI fails fast with a clear
missing meta credentials: run \muse login` or set META_API_KEY` message, so there's no silent breakage.Follow-ups deliberately left out
harness run --modelwon't override the model forsh -cpresets (no literal--modeltoken), as you noted in Add more agents to[tool.harness.agents]#49 — same as the other wrapped presets.harness configure-agentscontainment rules only cover Claude/Codex today; out of scope for this preset PR.Thanks for maintaining this — happy to adjust anything.