Rocq Beam

Rocq Beam

Use this skill when a Rocq project needs the narrow Rocq-facing surface that lean-beam already exposes, especially while porting Rocq developments to Lean. This is an optional auxiliary mode of the lean-beam toolchain, not a separate product or a second full workflow stack.

The goal is cheap Rocq proof-state inspection through coq-lsp, without turning Rocq support into a broad standalone interface.

Do not use coqtop or any fallback executor. Only coq-lsp is trusted. This is the Rocq-only skill. It should stay focused on Rocq and should not require Lean-specific workflow guidance. Do not factor shared Lean/Rocq skill instructions into a common helper; duplicate short guidance if both skills need it.

Setup

From the lean-beam repo root:

./scripts/install-beam.sh --codex --rocq-skill

Use --claude --rocq-skill, --pi --rocq-skill, --opencode --rocq-skill, or --vibe --rocq-skill instead when installing for Claude Code, Pi Agent, OpenCode, or Mistral Vibe. Use --all-skills --rocq-skill when you want every supported agent skill target.

The installer puts lean-beam in ~/.local/bin, stages the self-contained runtime under BEAM_INSTALL_ROOT (default ~/.local/share/beam), and installs the optional Rocq skill only when --rocq-skill is paired with a selected agent skill target.

Restart active agent sessions after installation.

The user-facing setup and installer reference is [docs/SETUP.md](../../docs/SETUP.md). The Rocq status and setup page is [docs/ROCQ.md](../../docs/ROCQ.md).

Rocq Setup

Rocq-specific setup:

cd /path/to/lean-beam
bash tests/setup-rocq-opam.sh

Skill Surface

This skill documents the current Rocq-facing lean-beam workflow surface. Keep the surface narrow: the current wrapper is for goal inspection against saved files, not for hidden proof-session mutation.

Supported command families:

• bootstrap the Rocq backend: lean-beam ensure rocq
• inspect goals after an existing sentence: lean-beam rocq-goals-after
• inspect goals before a sentence or after speculative sentence text within that basis:

lean-beam rocq-goals-prev

• inspect tracked files and daemon state: lean-beam open-files, lean-beam stats, lean-beam reset-stats

What to treat as the current agent workflow surface:

• default command: lean-beam rocq-goals-after
• intermediate-state command: lean-beam rocq-goals-prev with extra text when needed
• operational introspection: lean-beam open-files, lean-beam stats, lean-beam reset-stats

Core workflow contract:

• use lean-beam, not raw JSON and not raw LSP
• save the .v file before every new probe after a real edit
• lean-beam only sees the on-disk file, not unsaved editor buffers
• treat as LSP-style coordinates for the saved file: line 0 is the first

line, character 0 is the first character position on that line, and on a truly empty line only character 0 is valid

• there is no Rocq sync command in the current wrapper
• there is no Rocq handle or continuation surface in the current wrapper
• there is no Rocq run-at command in the current wrapper; use the goal probes instead
• do not assume hidden mutable proof-session state carries across requests
• do not use coqtop or a fallback executor; only coq-lsp is trusted

Use lean-beam, not raw JSON and not raw LSP.

lean-beam for Rocq:

• infers the target project root from the current directory or --root
• keeps one Beam daemon per project root and records it in /.beam/beam-daemon.json
• in sandboxed or read-only project trees, set BEAM_CONTROL_DIR to a writable directory
• owns Beam daemon startup, shutdown, and registry handling
• resolves coq-lsp from the target project's local _opam when available
• starts a Rocq-capable Beam daemon with explicit startup args instead of relying on inherited editor state
• wrapper commands talk to the per-project Beam daemon over localhost TCP; they are not direct in-process Rocq calls
• in Codex-style sandboxes, Beam daemon startup may still require elevated permissions even when all paths resolve correctly
• in the same environments, localhost TCP bind/connect for the Beam daemon and client may also require elevated permissions
• if startup fails with operation not permitted, treat that as a sandbox capability problem first, not as a missing install
• lean-beam shutdown, lean-beam stats, and lean-beam reset-stats apply to the current project only

Default rules:

• use lean-beam, not raw JSON and not raw LSP
• start with lean-beam rocq-goals-after
• save the file before every new probe after a real edit
• keep coordinates 0-based; do not guess editor-specific 1-based lines or columns
• if you think you want a Rocq run-at, use lean-beam rocq-goals-prev with extra text or rocq-goals-after instead
• use lean-beam rocq-goals-prev plus text when you need an intermediate state inside a sentence
• do not assume any hidden proof-session state carries across requests

Workflow

Ensure the Rocq backend:

lean-beam ensure rocq
lean-beam stats

Inspect goals after a sentence:

lean-beam rocq-goals-after "Demo.v" 2 8

Inspect goals before a sentence:

lean-beam rocq-goals-prev "Demo.v" 2 8

For a tactic sentence like a; b, inspect the intermediate state after a with:

lean-beam rocq-goals-prev "Demo.v" 2 8 "a."

Source-file model:

• lean-beam rocq-goals-* does not edit Demo.v
• edit the file normally, save it, then probe again
• lean-beam only sees the on-disk Rocq file, not unsaved editor buffers
• actual source edits happen through the normal file-edit workflow
• there is no Rocq sync command in the current wrapper; saving the file is the important step before the next probe

Execution model:

• every lean-beam rocq-goals-* request is an isolated read-only probe against the current saved file
• do not expect hidden mutable proof-session state to carry from one probe to the next
• the Beam daemon may reopen or resync the on-disk file before a probe, but saving the file is still the real boundary you control
• there is no Rocq lean-beam sync equivalent in the wrapper, so after edits the important step is: save, then probe again
• if the file changes while a request is pending or coq-lsp state becomes stale, expect to rerun from the saved file instead of relying on recovery inside the old request

Default loop:

lean-beam ensure rocq
lean-beam rocq-goals-after "Demo.v" 12 4

# make a real edit, save the file
lean-beam rocq-goals-after "Demo.v" 12 4

Use cases:

1. Inspect the current proof state after a sentence

lean-beam ensure rocq
lean-beam rocq-goals-after "Demo.v" 12 4

1. Inspect an intermediate tactic state inside one sentence

lean-beam ensure rocq
lean-beam rocq-goals-prev "Demo.v" 12 4 "intro x."
lean-beam rocq-goals-prev "Demo.v" 12 4 "split."

1. Check the effect of a small real edit

Save the file first, then probe again from the saved document.

lean-beam ensure rocq
lean-beam rocq-goals-after "Demo.v" 12 4

# make a real edit in Demo.v and save it
lean-beam rocq-goals-after "Demo.v" 12 4

Policy

• default to lean-beam rocq-goals-after
• use lean-beam rocq-goals-prev plus text for intermediate-state probing
• keep ppFormat as Str
• do not treat lean-beam as a source editor; actual .v edits happen through the normal file-edit workflow
• do not assume one goal probe mutates the basis of the next probe; each request starts from the current saved document state
• if coq-lsp reports stale or broken state unexpectedly, stop and report it loudly

Stats

Use:

lean-beam open-files
lean-beam stats
lean-beam reset-stats

lean-beam open-files shows the files currently tracked by the Beam daemon for the current project. For tracked files the broker already knows about, the wrapper checks status incrementally against the current on-disk text, and open-files also reports the last compact fileProgress observed for that tracked version.

Stats are in-memory only and scoped to the current project Beam daemon.

Upstream Rocq Features Not Yet Wrapped

Useful petanque/* methods we may expose later:

• petanque/get_state_at_pos
• petanque/run_at_pos
• petanque/goals
• petanque/premises
• petanque/ast_at_pos
• petanque/list_notations_in_statement
• petanque/proof_info_at_pos

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: ejgallego
• Source: ejgallego/lean-beam
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.