Stop Crediting an until read Loop as Bounded in the Guard

[ptr727]

Requirement 7's read bound credited a loop whose condition is a leading read over a descriptor-0 redirect without telling while from until. For until, an exhausted or empty input fails read forever, which keeps the condition false forever, so the loop never ends.

• _reads_its_input now takes the loop keyword and credits the bound for a while loop only. The while case is unchanged.
• The self-test carries until read l; do sleep 30; done < f as a deny beside the existing while read l; do sleep 30; done < f allow. Reverting the keyword check makes that case fail.
• _reads_its_input's docstring, the requirement 7 paragraph in host-setup/agent-safety/README.md, and its flowchart node now say the bound holds for a while loop.

A local strict review pass also found pre-existing ways a while read condition can fail to exhaust its input, filed as #2160 rather than fixed here.

Closes on promotion: #1633

🤖 Generated with Claude Code

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Repository: ptr727/ProjectTemplate/.coderabbit.yaml

Review profile: CHILL

Plan: Advanced

Run ID: 363a0e7e-c984-4de7-8c85-76ad5a761263

---

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.}

[Copilot]

Copilot review overview

🟢 Approval recommended

The change is minimal and fully verified — the keyword gate correctly denies until read loops, the sole caller passes the loop keyword, and a matching self-test plus consistent documentation updates accompany it.

Review effort: Balanced
Findings: None

What changed in this PR

This PR fixes a security-hardening defect in the Claude Code gh-write-guard.py hook that enforces requirement 7 (deny unbounded shell wait loops). Previously, the "read bound" exemption credited any loop whose condition began with read over a descriptor-0 redirect as bounded, without distinguishing while from until. For an until loop, an exhausted or empty input makes read fail forever, keeping the condition false forever, so the loop never exits — the exact opposite of a bound. The guard now grants the read-bound exemption only to while loops, and the accompanying documentation is corrected to match.

Changes:

• _reads_its_input gains a keyword parameter and returns False for any loop keyword other than while; the single caller now passes the loop keyword token.
• A new self-test case denies until read l; do sleep 30; done < f, beside the existing while read allow case, so reverting the keyword check fails the test.
• The docstring, README requirement 7 paragraph, and the flowchart node are updated to state the bound holds only for a while loop.

I verified that the caller passes the actual loop keyword (tok = toks[i] where _opens_loop is true), that _reads_its_input has no other callers, that the new tuple lands in _WAIT_CASES (exercised through classify), and that the arithmetic-for path is unaffected since its condition never begins with read. The prose changes are ASCII-clean with no banned issue references or mid-sentence semicolons. A related pre-existing weakness in the while read path was correctly filed separately rather than expanded into this change.

File	Description
host-setup/​agent-safety/​claude/​gh-write-guard.py	Adds while-only gate to _reads_its_input, updates the caller to pass the loop keyword, refreshes the docstring, and adds an until read deny self-test.
host-setup/​agent-safety/​README.md	Corrects requirement 7 prose and the flowchart node to state the read bound applies only to a while loop.

---

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

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (develop@e0ab4ee). Learn more about missing BASE report.

Additional details and impacted files

@@            Coverage Diff             @@
##             develop    #2161   +/-   ##
==========================================
  Coverage           ?   56.47%           
==========================================
  Files              ?       16           
  Lines              ?     7455           
  Branches           ?        0           
==========================================
  Hits               ?     4210           
  Misses             ?     3245           
  Partials           ?        0           

Flag	Coverage Δ
python-3.13	56.47% <ø> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.