file_read find_files() traverses hidden directories (.venv, .git, etc.) causing token explosion

[ranyakhemiri]

Bug

When file_read is called with a directory path, find_files() uses os.walk() and filters hidden files (file.startswith(".")) but does not filter hidden directories. This means os.walk descends into .venv, .git, __pycache__, node_modules, etc.

Impact

A .venv directory (~156MB) caused the tool to collect ~39M tokens worth of file content, crashing the agent with:

prompt is too long: 1069752 tokens > 200000 maximum

Reproduction

1. Create a project with a .venv directory (any standard Python venv)
2. Call file_read(path=".", mode="find") or file_read(path="data/", mode="view") where the agent resolves to a directory containing .venv
3. The tool walks into .venv and returns thousands of files

Root Cause

In find_files(), the directory branch:

for root, _dirs, files in os.walk(pattern):
    if not recursive and root != pattern:
        continue
    for file in sorted(files):
        if not file.startswith("."):  # Skip hidden files
            matching_files.append(os.path.join(root, file))

Hidden files are skipped, but _dirs is never modified in-place. Per Python docs, modifying dirnames in-place is the standard way to prune os.walk traversal.

Possible Fixes

Several approaches could work — leaving it to maintainers to decide:

• Minimal: Apply the same hidden-item logic to directories: dirs[:] = [d for d in dirs if not d.startswith(".")] — this alone catches .venv, .git, .mypy_cache, etc.
• Configurable: Add a FILE_READ_SKIP_DIRS env var (consistent with existing FILE_READ_RECURSIVE_DEFAULT pattern)
• Gitignore-aware: Use .gitignore rules to determine which directories to skip
• Guard: Add a max file count or total size limit to prevent runaway traversals

Environment

• strands-tools: latest (main)
• Python 3.12
• macOS

[dosvk]

@ranyakhemiri Opened a fix in #440. Filters hidden directories (starting with .) in place during os.walk.