chore(#5721): scope ty pre-commit hook to changed files

[fullsend-ai-coder]

Summary

• Remove . positional argument from the ty pre-commit hook entry so it no longer checks the entire project tree
• Drop pass_filenames: false so pre-commit appends only the staged Python files to ty check

Why

The hook was configured to run ty check . on every commit regardless of which files changed. This made commits slower and surfaced diagnostics unrelated to the staged files. Since ty check accepts individual file paths, letting pre-commit pass filenames is the correct behavior.

Testing

• Verified YAML syntax is valid after the change
• The fix is a two-line config change with well-defined behavior per the pre-commit framework and ty CLI documentation

---

Closes #5721

Post-script verification

• Branch is not main/master (agent/5721-ty-hook-pass-filenames)
• Secret scan passed (gitleaks — 8ac8468aeabcc94dc3be41d911c7fa97508211db..HEAD)
• PR body secret scan passed (gitleaks — no-git)
• Pre-commit hooks passed (authoritative run on runner)
• Tests ran inside sandbox

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 3:07 PM UTC · Completed 3:16 PM UTC
Commit: f580f56 · View workflow run →

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[fullsend-ai-review]

Review

Findings

Medium

• [protected-path] .pre-commit-config.yaml — This PR modifies .pre-commit-config.yaml, which is a protected governance/infrastructure file. The change is authorized by linked issue pre-commit: ty hook checks entire project instead of changed files #5721 and the PR description explains the rationale. Human approval is always required for protected-path changes, regardless of context.

---

Labels: PR modifies pre-commit config (CI infrastructure)

[fullsend-ai-retro]

🤖 Finished Retro · ✅ Success · Started 3:32 PM UTC · Completed 3:46 PM UTC
Commit: f580f56 · View workflow run →

[fullsend-ai-retro]

Retro: PR #5728 — scope ty pre-commit hook to changed files

Verdict: Clean workflow, no new proposals warranted.

This was a well-executed end-to-end pipeline for a simple 2-line config change to .pre-commit-config.yaml (removing . from the ty check entry and dropping pass_filenames: false).

Timeline

Stage	Run ID	Duration	Cost	Outcome
Triage	30463237700	~5 min	$0.45	Classified as ready-to-code (clarity 0.985)
Code	30463662911	~4.5 min	$1.24	Correct fix on first attempt, 28 tool calls
Review	30464182781	~9.5 min	$1.77	Protected-path finding (correct), comment verdict
Human	—	~8 min	—	Approved by rh-hemartin

Total: ~34 min issue-to-merge, ~$3.46, zero rework iterations.

Agents repo: fullsend-ai/agents at commit 8b9e25e.

What went well

• Code agent efficiency: 28 tool calls, single Edit call with the exact correct fix. No scope creep, graceful recovery from sandbox constraints (HTTP 403 on git fetch, network-blocked pre-commit run).
• Triage accuracy: Correctly classified a well-written issue as ready-to-code with appropriate labels (type/chore, component/ci, priority/low, good-fullsend-issue).
• Review correctness: The protected-path finding was well-calibrated — medium severity, non-actionable, correctly noting issue pre-commit: ty hook checks entire project instead of changed files #5721 authorization. The requires-manual-review label ensured human oversight for governance infrastructure.
• Pipeline velocity: Handoffs were fast — 5s triage→code, 12s code→review. Human approved 8 minutes after review posted.

Evidence supporting existing issues (not filing new proposals)

Review proportionality (agents#301, agents#439, agents#497): This run is another data point — $1.77 and 46 tool calls across 4 sub-agents for a 2-line YAML change that produced zero code findings. Only the orchestrator's structural protected-path flag was meaningful. The review sub-agents spent effort on tangential exploration: correctness globbed **/*.py to understand the Python file landscape (unnecessary for evaluating a config change), intent-coherence read ADR-0056 about the tools registry (irrelevant to hook configuration), and style-conventions made 7 reads/greps of .pre-commit-config.yaml at different offsets. A trivial-change fast path (agents#301) would have eliminated ~$1.50 of this cost.

Sub-agent redundant reads (agents#140, agents#252): Despite PR #172 (merged Jul 17) reportedly fixing cross-agent file duplication, the style-conventions sub-agent still made 7 tool calls reading the same file at different offsets. This may be intra-agent fragmentation rather than inter-agent duplication — a distinct pattern from what #140 addressed — but it contributes to the same cost overhead.

Triage output directory (fullsend#3012): The triage agent needed 4 attempts to write its output (directory didn't exist, then tried unavailable Write tool). Pre-creating FULLSEND_OUTPUT_DIR would save 2 tool calls per triage run.