registry / opencode-skills-collection / 4.0.71
opencode-skills-collection@4.0.71
OpenCode CLI plugin that automatically downloads and keeps skills up to date.
AI Security Review
scanned 1mo ago · by lpm-firewall-aiThis is first-party OpenCode plugin setup: on plugin load it mutates only the package-owned OpenCode skill vault and pointer directories, with no registry install hook or outbound channel. That guarded agent-surface write is a warning, and the inspected runtime does not show a separate theft, remote-execution, or install-hook attack.
Static reason
High-risk behavior combination matched malicious policy.
Trigger
OpenCode invokes the default plugin export at startup.
Impact
The OpenCode agent skill surface is populated and pruned to match this package release, including skills the default filter does not exclude.
Mechanism
runSkillPointer creates ~/.config/opencode/skills and ~/.config/opencode/skill-libraries, copies bundled skill directories, applies optional local regex patches, and rewrites pointer SKILL.md files.
AI rationale
This is first-party OpenCode plugin setup: on plugin load it mutates only the package-owned OpenCode skill vault and pointer directories, with no registry install hook or outbound channel. That guarded agent-surface write is a warning, and the inspected runtime does not show a separate theft, remote-execution, or install-hook attack.
Evidence
dist/index.jsdist/skill-pointer/pointer-generator.jsdist/skill-pointer/vault-installer.jsdist/skill-pointer/config-loader.jspackage.json~/.config/opencode/skills~/.config/opencode/skill-libraries~/.config/opencode/skill-filter.jsonc
Decision evidence
public snapshotAI called this Suspicious at 74.0% confidence as Dangerous Capability with medium false-positive risk.
Evidence for warning
- The published plugin entry copies bundled skills into the user OpenCode config the moment the plugin function runs.
- Active skill pointers are written under the home directory path .config/opencode/skills, and stale pointer directories there are deleted.
- The vault installer removes skill folders that are not in the current index, then recursively copies each bundled skill tree into ~/.config/opencode/skill-libraries.
- Default filter config excludes no risk levels, so bundled skills marked offensive or critical are installed unless the user adds a filter file.
- package.json scripts are limited to build, test, and prepublishOnly.
Evidence against
- There is no preinstall, install, or postinstall hook.
- dist contains no child_process, eval, Function, dynamic import, or HTTP client.
- Copy and patch paths reject parent-directory segments and confine patches inside the vault real path.
- Cleanup in the active skills directory is limited to folders whose names end with -category-pointer.
Behavioral surface
ChildProcessCryptoDynamicRequireEnvironmentVarsFilesystemNetworkShell
HighEntropyStringsUrlStrings
WildcardDependency
Source & flagged code
10 flagged · loading sourcebundled-skills/developer-signup-flow/SKILL.mdView file
165patternName = stripe_test_secret
severity = high
line = 165
High
High Secret
Package contains a high-severity secret pattern.
bundled-skills/developer-signup-flow/SKILL.mdView on unpkg · L165165patternName = stripe_test_secret
severity = high
line = 165
High
Secret Pattern
Stripe test secret key in bundled-skills/developer-signup-flow/SKILL.md
bundled-skills/developer-signup-flow/SKILL.mdView on unpkg · L165bundled-skills/playwright-skill/run.jsView file
12L13: const fs = require('fs');
L14: const path = require('path');
Medium
Dynamic Require
Package source references dynamic require/import behavior.
bundled-skills/playwright-skill/run.jsView on unpkg · L12bundled-skills/apify-content-analytics/reference/scripts/run_actor.jsView file
98const apiActorId = actorId.replace('/', '~');
L99: const url = `https://api.apify.com/v2/acts/${apiActorId}/runs?token=${encodeURIComponent(token)}`;
L100:
...
L102: try {
L103: data = JSON.parse(inputJson);
L104: } catch (e) {
...
L114: },
L115: body: JSON.stringify(data),
L116: });
...
L321: // Check for APIFY_TOKEN
L322: const token = process.env.APIFY_TOKEN;
L323: if (!token) {
Critical
Hardcoded Runtime Data Exfiltration
Source sends credentials or rich application records to a package-controlled external receiver enabled by default.
bundled-skills/apify-content-analytics/reference/scripts/run_actor.jsView on unpkg · L98bundled-skills/huggingface-tool-builder/references/baseline_hf_api.tsxView file
39L40: const token = process.env.HF_TOKEN;
L41: const headers: Record<string, string> = token
...
L44:
L45: const url = `https://huggingface.co/api/models?limit=${limit}`;
L46:
...
L54:
L55: const text = await res.text();
L56: process.stdout.write(text);
L57: })();
High
Credential Exfiltration
Source combines credential-like environment material and outbound requests; review data flow before blocking.
bundled-skills/huggingface-tool-builder/references/baseline_hf_api.tsxView on unpkg · L39bundled-skills/gemini-deep-research/scripts/research.pyView file
•path = bundled-skills/gemini-deep-research/scripts/research.py
kind = build_helper
sizeBytes = 24204
magicHex = [redacted]
Medium
Ships Build Helper
Package ships non-JavaScript build or shell helper files.
bundled-skills/gemini-deep-research/scripts/research.pyView on unpkgbundled-skills/web-artifacts-builder/scripts/shadcn-components.tar.gzView file
•path = bundled-skills/web-artifacts-builder/scripts/shadcn-components.tar.gz
kind = compressed_blob
sizeBytes = 19967
magicHex = [redacted]
Medium
Ships Compressed Blob
Package ships compressed or archive-like blobs.
bundled-skills/web-artifacts-builder/scripts/shadcn-components.tar.gzView on unpkgbundled-skills/theme-factory/theme-showcase.pdfView file
•path = bundled-skills/theme-factory/theme-showcase.pdf
kind = high_entropy_blob
sizeBytes = 124310
magicHex = [redacted]
High
Ships High Entropy Blob
Package ships high-entropy non-source blobs.
bundled-skills/theme-factory/theme-showcase.pdfView on unpkgbundled-skills/fedora-hyprland-installer/tests/test-scripts.shView file
•path = bundled-skills/fedora-hyprland-installer/tests/test-scripts.sh
kind = payload_in_excluded_dir
sizeBytes = 1740
magicHex = [redacted]
High
Payload In Excluded Dir
Package hides binary, compressed, or executable-looking payloads in test/fixture/hidden paths.
bundled-skills/fedora-hyprland-installer/tests/test-scripts.shView on unpkgbundled-skills/go-in-depth/scripts/workflow-script.jsView file
•stage = ast_semantic_analysis; reason = ast_parse_error; limitedFiles = 2
High
Semantic Analysis Limited
A bounded semantic-analysis stage reached its safety limit; remaining detectors completed, but this package requires AI review.
bundled-skills/go-in-depth/scripts/workflow-script.jsView on unpkgFindings
1 Critical6 High7 Medium5 Low
CriticalHardcoded Runtime Data Exfiltrationbundled-skills/apify-content-analytics/reference/scripts/run_actor.js
HighHigh Secretbundled-skills/developer-signup-flow/SKILL.md
HighCredential Exfiltrationbundled-skills/huggingface-tool-builder/references/baseline_hf_api.tsx
HighShips High Entropy Blobbundled-skills/theme-factory/theme-showcase.pdf
HighPayload In Excluded Dirbundled-skills/fedora-hyprland-installer/tests/test-scripts.sh
HighSemantic Analysis Limitedbundled-skills/go-in-depth/scripts/workflow-script.js
HighSecret Patternbundled-skills/developer-signup-flow/SKILL.md
MediumDynamic Requirebundled-skills/playwright-skill/run.js
MediumNetwork
MediumEnvironment Vars
MediumShips Build Helperbundled-skills/gemini-deep-research/scripts/research.py
MediumShips Compressed Blobbundled-skills/web-artifacts-builder/scripts/shadcn-components.tar.gz
MediumStructural Risk Force Deep Review
MediumWildcard Dependency
LowNon Install Lifecycle Scripts
LowScripts Present
LowFilesystem
LowHigh Entropy Strings
LowUrl Strings
Affected versions and remediation
This report applies to opencode-skills-collection@4.0.71.
See all versions of opencode-skills-collection for other recorded verdicts.
- Review the evidence and your use of opencode-skills-collection@4.0.71 before allowing it. Restrict the permissions described in this report.
- Choose an independently verified alternative or release. This report does not establish that other versions are safe.
Evidence last updated: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.