Ghostclaw

PassAudited by ClawScan on May 1, 2026.

Overview

Ghostclaw appears purpose-aligned for code architecture review, but its optional background monitoring, GitHub PR automation, credential use, and reporting features should be enabled carefully.

Ghostclaw is not showing artifact-backed malicious behavior. Before installing, decide whether you actually want background scans, PR automation, self-updates, caching, or Telegram notifications. If you use GitHub automation, use least-privilege credentials and keep the repository list narrowly scoped.

Findings (7)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

If enabled, Ghostclaw can create visible branches, commits, and pull requests in repositories.

Why it was flagged

When --create-pr is used, the tool creates a branch, commits a report, pushes it, and opens a GitHub PR. This is disclosed and purpose-aligned, but it mutates repositories and GitHub state.

Skill content
subprocess.run(["git", "checkout", "-b", branch_name]...); subprocess.run(["git", "commit", "-m", ...]); subprocess.run(["git", "push", "origin", branch_name]...); pr_cmd = ["gh", "pr", "create", "--title", title, "--body", body]
Recommendation

Use --create-pr only for repositories where this is acceptable, review generated reports before publishing, and prefer dry-run or no-write modes when testing.

What this means

If enabled, Ghostclaw may keep scanning projects and producing feedback or PRs without a fresh manual invocation each time.

Why it was flagged

The hook mode is a disclosed persistent/background mode that can act on command and system events. It is not shown as hidden, but it is a higher-trust operating mode.

Skill content
Ghostclaw operates in the background... Listens to Events... periodic system heartbeats... can silently log its findings or proactively open PRs if issues cross a threshold.
Recommendation

Enable hook or cron mode only for specific repositories, keep autoCreatePRs disabled unless intended, and know how to disable the hook.

What this means

A broadly scoped GitHub token could allow repository changes beyond what the user intended for architecture reporting.

Why it was flagged

GitHub credentials are expected for PR automation, but they give the skill delegated authority over repositories.

Skill content
Set `GH_TOKEN` env variable for PR automation
Recommendation

Use a fine-grained token with the minimum repository permissions needed, and avoid sharing broad personal access tokens with background watcher mode.

What this means

PR and notification behavior may depend on code not present in the supplied artifacts or may fail if the package is incomplete.

Why it was flagged

Watcher mode references helper modules for cache, GitHub, and notifications that are not listed in the provided file manifest, while the registry also has no install spec. This looks more like incomplete packaging than malicious behavior, but users should verify the package source before using those features.

Skill content
from ghostclaw.lib.cache import VibeCache
from ghostclaw.lib.github import GitHubClient
from ghostclaw.lib.notify import Notifier
Recommendation

Install only from a trusted source and verify the complete package contents, especially the GitHub and notification helper modules, before enabling automation.

What this means

Running the update option can execute package installation/update logic from external sources.

Why it was flagged

The self-update path can run git pull or pip install. It is user-invoked and disclosed, but it can replace code with a newer package version.

Skill content
parser.add_argument("--update", action="store_true", help="Self-update Ghostclaw via pip or git") ... subprocess.run([sys.executable, "-m", "pip", "install", "--upgrade", "ghostclaw"], check=True)
Recommendation

Avoid --update unless you trust the package source and prefer reviewing updates through your normal dependency-management process.

What this means

Architecture findings and repository metadata may remain on disk after analysis.

Why it was flagged

The tool persists analysis reports locally. This is expected for trend analysis and performance, but reports can reveal repository structure and issue summaries.

Skill content
This cache stores full analysis reports keyed by a fast repository fingerprint... self.cache_dir = Path.home() / ".cache" / "ghostclaw"
Recommendation

Use a custom cache directory or clear the cache for sensitive repositories.

What this means

Project names and architectural issue summaries could be shared outside the local environment.

Why it was flagged

Notification mode can send repository names, scores, issue summaries, and PR URLs to an external messaging service when enabled.

Skill content
parser.add_argument("--telegram-token", help="Telegram bot token") ... notifier.notify(repo=repo_url, vibe_score=vibe_score, delta=delta, issues=issues[:10], pr_url=...)
Recommendation

Enable notifications only for approved channels and avoid sending sensitive repository information to untrusted chats.