技能 2026-06-09 01:00 UTC
claude-brain
由提交人 omonien 提交,主要更新类型为:问题缺陷修复与稳定性提升。提交说明:fix(import): handle keybindings.json shape mismatch without crashing (#44)。解决了已知运行时异常与边界错误,提升系统稳定性。
Commit提交信息原文
fix(import): handle keybindings.json shape mismatch without crashing (#44)
* fix(import): handle keybindings.json shape mismatch without crashing
Claude Code's keybindings.json may be either an array of bindings or
an empty `{}` object (when no bindings are configured). The current
union-merge:
jq -s '.[0] + .[1] | unique_by(.key, .command)' \
"${CLAUDE_DIR}/keybindings.json" "$tmp_remote_kb" > "$tmp"
requires both inputs to be arrays. When either side is an object, jq
exits with:
jq: error: object ({}) and array ([]) cannot be sorted,
as they are not both arrays
Because import.sh runs under `set -euo pipefail`, this crash terminates
import_brain mid-way — the shared-namespace import that follows and the
"Brain import complete." marker never run. The user sees the import
"succeed" (settings, MCP servers, etc. did land) but everything after
keybindings silently drops on the floor.
Fix
---
Detect both shapes before merging:
* array + array → union via unique_by (existing behaviour)
* array remote + non-array local → adopt remote
* non-array remote → log and skip; nothing to merge
Add an integration test (test_keybindings_shape_mismatch) that builds
a brain whose keybindings.content is `{}`, seeds local keybindings.json
as `{}` too, and asserts:
1. the import does not emit the "object and array cannot be sorted"
jq error, AND
2. "Brain import complete." still appears in the log — proof that
import_brain ran to the end instead of dying at the keybindings step.
Without the fix the new test fails as expected.
* review fixes: atomic write + early-return on first failure
Address Copilot review comments on #44.
1. Replace the two new `echo "$new_keybindings" > <file>` writes with
`printf '%s\n' "$x" > $tmp && mv $tmp <file>`. Same atomic-write
pattern the merge branch already uses. Avoids the `echo -n` /
non-atomic / escape-handling edge cases the reviewer flagged.
2. In `test_keybindings_shape_mismatch`, return after the first `fail`.
When the underlying bug fires, both assertions would fail (crash text
present AND "Brain import complete" missing) and the test would
report two correlated failures for one root cause. Now reports a
single clear failure per run.