Back to skill

Security audit

Sol Build Session

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a disclosed productivity workflow, but it tells autonomous agents to commit and push repository changes without review or user approval.

Install only if you are comfortable with autonomous build sessions, and override the commit step so the agent reviews diffs, stages only intended files, checks for secrets, and asks before pushing to any remote. Treat the logging helper as persistent local memory storage for whatever title and insight you pass to it.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:47
Finding
Unrestricted Staging and Remote Publication of Repository Content## Vulnerability Details **File Location**: `SKILL.md`, lines 47–51 **Vulnerability Type**: Unsafe autonomous Git staging and publication **Risk Level**: High ```bash If you wrote code: ```bash git add -A git commit -m "descriptive message" git push ``` ``` ### Technical Analysis The skill instructs an autonomous agent to run `git add -A`, commit, and push whenever it writes code. The `git add -A` command stages all tracked modifications, deletions, and untracked files in the repository rather than limiting the commit to the intended deliverable. No review of the staged diff, secret scanning, file allowlist, branch validation, remote validation, or user confirmation is required before `git push`. Consequently, unrelated work, generated artifacts, private data, or credentials not protected by `.gitignore` can be included in the commit and transmitted to the repository's configured remote. ### Attack Path 1. A sensitive, unrelated, or attacker-controlled file exists or is introduced anywhere in the repository. 2. An autonomous build session writes code and follows the skill's commit procedure. 3. `git add -A` stages the intended work together with every other repository change and untracked file. 4. The agent commits the complete staged set without inspecting it. 5. `git push` transmits the commit to the configured remote. 6. Anyone with access to that remote may retrieve the unintentionally published content; removing it from the latest revision may not remove it from Git history. ### Impact Assessment Exploitation does not directly grant additional local operating-system privileges. Its impact is on the confidentiality and integrity of the repository and its remote: - Disclosure of credentials, tokens, private source code, memory files, or other sensitive artifacts. - Publication of unrelated or incomplete work. - Unauthorized modification of a shared remote branch within the agent's existing Git crede ...[truncated 362 chars]
Remediation
## Remediation Suggestions - Remove unconditional autonomous `git push` guidance and require explicit user approval before remote publication. - Stage only named deliverable paths, for example `git add -- path/to/file`, rather than using `git add -A`. - Inspect the staged content with `git diff --cached --stat` and `git diff --cached` before committing. - Run an appropriate secret scanner against staged content and block commits containing credentials or sensitive material. - Verify the active branch and remote URL before committing or pushing. - Ensure `.gitignore` excludes local secrets, memory data, generated files, and environment-specific artifacts, while recognizing that `.gitignore` is not a substitute for review. - Use protected branches and least-privilege Git credentials that cannot bypass review controls. - Prefer creating a local commit or patch and presenting it for review instead of automatically publishing it. A safer workflow is: ```bash git add -- path/to/intended-deliverable git diff --cached --stat git diff --cached # Run a secret scan and obtain user approval. git commit -m "descriptive message" # Push only after explicit approval and remote/branch verification. ```
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Vague Triggers

Medium
Confidence
94% confidence
Finding
The skill's activation criteria are very broad, covering generic autonomous time, cron sessions, and any opportunity to 'produce something.' That increases the chance the agent will invoke the skill in inappropriate contexts and take self-directed actions without clear user intent or task scoping, which can lead to unnecessary modifications or actions.

Missing User Warnings

Medium
Confidence
96% confidence
Finding
The skill explicitly instructs the agent to run git add, git commit, and git push as a default follow-up to writing code, with no requirement for user confirmation or review. This is dangerous because it can modify repository history and publish changes to remote infrastructure automatically, potentially leaking sensitive work, pushing broken code, or making unauthorized changes.

Missing User Warnings

Low
Confidence
82% confidence
Finding
This code creates a dated markdown file and appends user-provided session content into it under the workspace memory directory. Although it prints the destination after writing, the user-facing usage/comments do not warn beforehand that invoking the script will persist supplied content to disk.

Static analysis

No suspicious patterns detected.