fix(strategy): keep non-ASCII staged paths literal so commit linking matches

[wangzhengzhuo05]

What

getStagedFiles in cmd/entire/cli/strategy/manual_commit_hooks.go now parses git diff --cached --name-only -z and splits the output on NUL, so staged paths are returned literally instead of as git's quoted/C-escaped display form.

Why

With git's default core.quotePath=true, the newline-delimited --name-only output quotes and C-escapes unusual pathnames — café.go comes back as the display string "caf\303\251.go". That string was exact-compared against the literal path stored in SessionState.FilesTouched, so a session that touched only a non-ASCII file was classified as contributing no content to the commit: the commit was still created, but it received no Entire-Checkpoint trailer and its provenance (entire explain attribution) was incomplete. Reported with a focused control/canary reproduction in #2398: an ASCII control file got its trailer, an otherwise equivalent UTF-8 canary did not.

How

• Pass -z to the git invocation and iterate the NUL-delimited fields, keeping filepath.ToSlash on every entry.
• Drop the \r\n → \n normalization: it only existed to unwrap the newline framing, and -z emits NUL-terminated, unquoted names.
• The function's contract is unchanged: (non-nil empty slice, nil) when nothing is staged, (nil, err) only when git itself fails, and the same error wrapping.
• This follows the precedent already in the same package: priorAICommitFiles in cmd/entire/cli/strategy/telemetry_signals.go uses -z for the same reason, so paths match their FilesTouched form.

Tests

New cmd/entire/cli/strategy/staged_files_quoting_test.go, TestGetStagedFiles_PreservesNonASCIIPaths: sets core.quotePath true explicitly, stages a literal café.go plus an ASCII control.go, and asserts both come back literal and that no returned entry is git-quoted or C-escaped.

$ gofmt -l cmd/entire/cli/strategy/manual_commit_hooks.go cmd/entire/cli/strategy/staged_files_quoting_test.go
(no output)
$ go test ./cmd/entire/cli/strategy/ -count=1
ok  	github.com/entireio/cli/cmd/entire/cli/strategy	25.187s
$ go build ./...
(exit 0)

Mutation check

Reverting only the fix (removing -z, i.e. the previous behaviour) makes the new test fail with exactly the string from the report — the assertion output is the single staged entry "\"caf\\303\\251.go\"\ncontrol.go\n", which contains neither literal filename. Restoring -z returns the package suite to green, so the test guards the defect rather than the implementation.

Fixes #2398.

[Copilot]

🟢 Approval recommended

No unresolved issues remain, and regression coverage validates the fix.

Pull request overview

Fixes commit-to-session linking for non-ASCII staged paths by preserving literal Git filenames.

Changes:

• Parses staged paths using Git’s NUL-delimited output.
• Adds regression coverage for non-ASCII filenames.

File summaries

File	Description
cmd/entire/cli/strategy/staged_files_quoting_test.go	Verifies literal non-ASCII and ASCII paths.
cmd/entire/cli/strategy/manual_commit_hooks.go	Uses NUL-delimited staged-path parsing.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.