fix(compiler): leave tail -n +N to the VFS shell instead of printing the last N lines

[breken-ai]

Summary

tail -n +N prints a file from line N onward. The compiled Bash fast path (src/hooks/bash-command-compiler.ts, parseHeadTailStage) read +N as N and returned the last N lines instead.

So when an agent pages a memory file with tail -n +3 ~/.deeplake/memory/summaries/…, the pre-tool-use hook answers from the compiled path with the wrong part of the file, and nothing tells the agent.

Reproduction on ce30de7, with a file holding l1…l8:

command	GNU tail / just-bash	compiled fast path before	after
tail -n +3 /a.md	l3 … l8	l6 l7 l8	declines, so the VFS shell answers with l3 … l8

Fix: parseHeadTailStage returns null for tail -n +N. The compiler then declines the command, and the VFS shell handles it. Its just-bash tail implements the +N form. I checked that tail -n +3 there prints l3…l8. The other fast paths in pre-tool-use.ts don't match +N (their regex is -?\d+), so nothing else picks up the command before the shell does. cat f | tail -n +N was already rejected by isValidPipelineHeadTailStage.

This PR adds a regression test in tests/claude-code/bash-command-compiler.test.ts. It fails on main because the parser returns a cat segment, and it passes with the fix.

Version Bump

Not bumped. This is a patch-level bug fix, so I've left the release decision to you.

Test plan

• Tests pass locally (npm test): 5823 passed. The 17 failures are the same on the untouched main checkout and come from the environment: cowork-queue-leak, install-cowork, skillify-state, plugin-cache-gc-bundle, and two graph suites.
• Relevant new tests added: does not compile tail -n +N…. It fails on main and passes on this branch.
• npm run typecheck and npm run dup pass.
• Version bumped in package.json, or no release needed for this change

An AI agent (breken-ai) found this and wrote the fix and test. I reviewed and ran everything above before opening the PR.

Summary by CodeRabbit

• Bug Fixes
  • tail -n +N commands now use shell handling to return lines from line N onward, rather than being interpreted as a last-N-lines window.

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository: activeloopai/hivemind/.coderabbit.yaml

Review profile: CHILL

Plan: Advanced

Run ID: a69644b5-5331-485f-8142-b43e764f8db9

📥 Commits

Reviewing files that changed from the base of the PR and between ce30de7 and 2ba92c1.

📒 Files selected for processing (2)

• src/hooks/bash-command-compiler.ts
• tests/claude-code/bash-command-compiler.test.ts

Included review availability: Your plan provides up to 4 included reviews per hour; 3 remain after this review.

---

📝 Walkthrough

Walkthrough

The bash command compiler now rejects tail -n +N commands. A test verifies that the compiled parser rejects the command and that compiled execution returns null without reading virtual-path contents.

Changes

Tail from-line handling

Layer / File(s)	Summary
Reject +N tail syntax src/hooks/bash-command-compiler.ts, tests/claude-code/bash-command-compiler.test.ts	parseHeadTailStage returns null for tail -n +N. The test checks that parsing and compiled execution return null, and that execution does not read virtual-path contents.

Priority: ➖ Normal

Estimated code review effort: 2 (Simple) | ~5 minutes

Change: Bug fix

Merge Risk: ⚪ Minimal · up to 2ba92

tail -n +N is routed to the virtual shell rather than producing the incorrect compiled output. No merge-blocking issue is established.

Architecture Summary

Architecture risk: 🔵 Low · up to 2ba92

The change affects 2 systems.

Changed systems: src, tests

Architecture concerns
No architecture-level concerns identified.

Review details

Systems and components

• observed — src (service) was modified; 1 changed file maps to changed impact.
• observed — tests (service) was modified; 1 changed file maps to changed impact.

Before / after behavior

• observed — Modified behavior in src/hooks/bash-command-compiler.ts: parseHeadTailStage now returns null for tail -n +N; previously this case continued into numeric parsing and could be represented as a last-N-lines window. The added comment states that the VFS shell handles the +N form.
• observed — Modified behavior in tests/claude-code/bash-command-compiler.test.ts: Added a test for tail -n +3 /a.md: it expects parseCompiledSegment to return null, and compiled execution to return null without reading virtual-path contents.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 1 functions across 2 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main change: deferring tail -n +N handling to the VFS shell instead of compiling it incorrectly.
Description check	✅ Passed	The description includes the required Summary, Version Bump, and Test plan sections. It explains the bug, fix, regression test, validation results, and the absence of a version bump. The release decis…
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

• Fix all pre-merge checks with AI

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create a new PR

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}