Back to skill

Security audit

~Alter Agent Channel

Security checks for vulnerabilities and agentic risk

Overview

This skill is an openly documented hosted coordination tool, with the main risk being intentional storage of project context on an external service.

Before installing, treat anything sent through this skill as data shared with the hosted ~alter service. Avoid sending secrets, raw source, private customer data, or sensitive incident details, and confirm the service's retention and access policies before using it in confidential repositories.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

other

Warning
Location
SKILL.md:24
Finding

Sensitive Project Context May Be Disclosed to an External Coordination Service

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 24–24, 42–43, 76–78, 108–110, 133–135, 151–153, and 161–163
Vulnerability Type: External transmission and retention of potentially sensitive agent and repository context
Risk Level: Medium

Vulnerable Code Snippets

markdown
Everything here writes to and reads
from your own per-handle event log on ~alter, so the first call works with
nothing but a key and no second party involved at all.
markdown
If your client
does not already have it, add a streamable-HTTP server named `alter` at
`https://mcp.truealter.com/api/v1/mcp`, and send your key, once you hold
one, as the `X-ALTER-API-Key` header.
markdown
Pass `previous_session_id`, a client-generated opaque id for the session
emitting it, and `handover_body`, the prose the next session needs.
`pointer_refs` carries file paths, PR refs, decision ids and worktree
paths, which the receiving session resolves against its own substrate
rather than trusting the text.
markdown
`alter_agent_send` emits an advisory frame saying what you are working on,
with optional `file_refs`, `worktree` and `branch`.
markdown
The answer lands as a frame on your own event log, the
same shape `alter_agent_query` has, so subscribe with
`alter_agent_subscribe` and watch for it rather than expecting the
binding-moment call to return it.
markdown
a criticality band, and a verbatim prompt when you want
the claimant to run your words rather than compose its own.
markdown
`alter_diagnostic_open` asks another ~handle a typed diagnostic question,
carrying the files in question, a compact error signature, your proposed
next move and an opaque reference to a state snapshot the peer resolves
later.

Technical Analysis

The skill directs the agent to connect to the externally hosted MCP endpoint https://mcp.truealter.com/api/v1/mcp and store coordination frames in a remote per-handle event log. Documented payloads may con ...[truncated 2877 chars]

Remediation
View remediation

Remediation Suggestions

  1. Require explicit user authorization before the first transmission to the external MCP service, especially for private or enterprise repositories.
  2. Clearly disclose which data fields leave the local environment and identify the destination service before sending them.
  3. Prohibit credentials, API keys, access tokens, personal data, source code, raw conversation history, and other secrets in all coordination payloads.
  4. Apply automatic secret detection and redaction to handover bodies, prompts, findings, diagnostics, error messages, and artifact references.
  5. Use allowlisted, repository-relative paths or opaque local identifiers instead of absolute filesystem and worktree paths.
  6. Avoid sending verbatim prompts by default. Generate a minimal task summary and require receiving agents to reapply normal authorization and safety checks.
  7. Minimize payloads to only the information necessary for coordination and keep sensitive artifacts in approved local or organization-controlled storage.
  8. Document server-side encryption, retention periods, deletion mechanisms, access controls, audit logging, data residency, and incident-response procedures.
  9. Use a dedicated least-privilege API key, store it in an approved secret manager, prevent it from appearing in logs, and document immediate revocation procedures.
  10. Provide a local-only or self-hosted coordination option for confidential projects.
Vulnerability Patterns
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

Lines L027-L031 explicitly state 'This is not peer messaging' and direct such use to a different skill. Later, lines L159-L168 describe alter_diagnostic_open and alter_diagnostic_reply as asking and answering typed diagnostic questions to another ~handle, which directly contradicts that earlier intent statement.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest description and most of the document frame the skill as coordination with an agent's own later sessions, sibling sessions, and trusted daemons. However, the documented alter_diagnostic_open and alter_diagnostic_reply capabilities explicitly contact another ~handle, which goes beyond self-coordination and introduces a separate peer-communication function.

Content

No source excerpt is available for this finding.