broca/.gitignore at master · cortexkit/broca · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
# Rust build output
/target
**/*.rs.bk

# ENGINE RUNTIME STATE. `broca-run` writes its WAL and single-writer leases under
# `<project-root>/.broca/`, and --project-root defaults to the CWD -- so running the CLI
# from a checkout puts session state inside the repo. That is correct behaviour for a
# harness whose durable log belongs beside the project it drives; what was wrong was
# tracking the result.
#
# A WAL is a TRANSCRIPT: prompts, tool arguments and tool results, verbatim. The two files
# this rule retired held only a `say pong` acceptance probe, but the same command run
# against real work would commit that work's contents to a public repository. Nothing about
# the file makes that visible in a diff -- it is opaque framed bytes -- so the review that
# would catch it does not exist.
/.broca/
**/.broca/

# Node deps for the e2e test harness (aimock) — reinstall with `npm ci` from the
# committed package.json + package-lock.json. The installed tree is not a deliverable.
/node_modules

# Commit Cargo.lock (this workspace builds binaries).

.DS_Store

# Agent runtime + local tooling installs (reinstallable, not project deliverables)
/.alfonso
/.prefrontal
/.opencode
/.agents
# CortexKit layout: internal docs, scripts and tool state stay out of git.
/.cortexkit/*
!/.cortexkit/release-notes/
!/.cortexkit/skills/
!/.cortexkit/rules/
!/.cortexkit/WORKER.md

# Auto-generated docs from overnight tooling (not hand-maintained deliverables;
# regenerated, and were repeatedly swept into commits by `git add -A`). Left on disk
# for reference, kept out of the repo.
/ARCHITECTURE.md
/STRUCTURE.md

# Bench capture artifacts (large, regenerable)
/crates/broca-bench/captures/*.sse

# (A rule for /crates/broca-provider/tests/live/captures/ was removed here. Those 22 files
# are tracked, nothing on disk reads them, and the rule's only effect was to hide a NEW
# family's capture from `git status`. The reasoning lives beside the writer, in
# crates/broca-provider/tests/live_common/mod.rs, which is where someone deciding whether
# to commit a capture is actually looking.)

# STAGED LOOP CORPUS -- NEVER COMMITTABLE.
#
# Rounds of THE LOOP stage a read-only extract here for a worker that cannot reach the live
# daemon: an export/run-index slice and a sample of REAL PRODUCTION WALs. Those WALs carry
# full transcripts -- user prompts, tool arguments, tool results -- and this repository has
# outside contributors.
#
# Without this rule the corpus sits untracked in a worktree, one `git add -A` away from
# being published. The rule is the guard; a convention to be careful is not.
#
# Consequence, and it is intended: the corpus is STAGED PER ROUND, never carried in the
# tree. A prompt that needs a reading gets one staged beside it.
.loop-corpus/

# Test-runner output. Three nextest logs were committed in a3dbb8c4 (2026-07-10) and
# tracked until a fleet gitignore census found them: 133 KB of PASS lines and cargo
# compile paths, benign, but a run's output is evidence of one run on one machine and
# not a fact about the tree. No .log file has ever been legitimately tracked here.
*.log

# Test-rig binaries staged into worker worktrees by the head (see .cortexkit/prefrontal/scripts,
# formerly .cortexkit/alfonso/scripts).
/.test-rigs/

# Python bytecode written when the script suites import helper modules.
__pycache__/
*.pyc