Back to skill

Security audit

canonry

Security checks for vulnerabilities and agentic risk

Overview

The skill documents legitimate Canonry workflows and requires specific approval before changes, paid requests, or recurring automation.

Use stored-only access for analysis and grant project-specific permissions when possible. Approve exact changes, spending, and recurring schedules separately; keep credentials out of chat. Traffic integrations send visitor metadata to your configured Canonry server, so review collection and retention before enabling them. Revoke dedicated credentials during teardown. The scanner's incomplete references were reviewed directly, but this assessment does not verify the separately installed runtime.

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
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (31)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The declared purpose describes inspection, analysis, planning, review, and reporting, but the commands provision credentials with advertising mutation and approval privileges. This is a materially additional access-management capability, not merely an implementation detail of the stated workflows. The commands do not themselves modify or approve ads, but they create keys authorized to do so.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The declared purpose describes inspection, analysis, planning, and integration review. The supplied commands instead make operational changes to Cloudflare infrastructure and enable data delivery to Canonry. These are material provisioning and deployment capabilities, not merely implementation details of reviewing an integration. The chunk contains reference commands rather than evidence of automatic execution, but the documented capabilities still exceed the declared scope.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The inspection, status, audit, and comparison commands fit the declared purpose. However, publishing content, modifying SEO configuration, deploying schema, and changing integration connections are substantive write and administration capabilities not conveyed by a description centered on inspection, diagnosis, planning, review, and reporting. Although this chunk is only a command reference and shows no execution, the workflows it exposes extend materially beyond the declared scope.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

Process status and logs can support the declared diagnostic workflows. However, pm2 restart canonry actively controls the service lifecycle rather than merely inspecting or reporting on it. The description does not disclose service restart or runtime administration capabilities.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

Key inspection could support authentication troubleshooting, but creating privileged credentials and revoking keys are security-administration capabilities not disclosed by the stated evidence and reporting workflows. The chunk documents these commands rather than demonstrating that they were executed; there is no evidence of credential exfiltration.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

Inspection and status commands align with the declared purpose. However, project lifecycle management and persistent configuration changes materially exceed the stated inspection, diagnosis, planning, review, and reporting activities. The deletion dry-run option does not remove this discrepancy because an actual deletion command is also documented.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

The declared purpose describes inspection, diagnosis, planning, review, and reporting. Requesting indexing is an externally consequential action rather than a supporting detail of those activities. Although the commands use the declared cnry CLI, the description does not disclose this indexing-submission capability.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
89% confidence
Finding

The description emphasizes inspection, diagnosis, planning, review, and reporting. The documented configuration changes and Google submission operations go beyond those activities and introduce material state-changing capabilities that are not disclosed. Synchronization and backfill are reasonable supporting details for evidence analysis. This chunk is a command reference, not proof that any commands were executed.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
88% confidence
Finding

Exporting project results and inspecting a sitemap fit the declared inspection and reporting workflows. However, cnry apply introduces configuration changes rather than merely inspecting, diagnosing, planning, or reporting. That state-changing capability is not represented in the declared purpose. The chunk presents command examples rather than evidence that these operations have been executed.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
88% confidence
Finding

Listing targets and dismissals fits the declared inspection purpose. Dismissing and restoring targets changes application state, rather than merely inspecting, planning, reviewing, or reporting. This workflow-management capability is not disclosed in the description.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
88% confidence
Finding

The declared purpose describes inspection, review, planning, and reporting, but the supplied commands change project configuration and establish or remove an external webhook destination. This is a substantive integration-management capability, not merely reviewing integrations. The chunk documents these commands; it does not itself demonstrate their execution or any actual data transmission.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
87% confidence
Finding

Evidence inspection, history queries, and ordinary exports support the declared purpose. However, secret-inclusive exports and state-changing backfills are materially broader capabilities not disclosed in the description. The dry-run variants do not make all backfills read-only because non-dry-run forms are also documented. The snippet does not itself execute these commands or demonstrate unauthorized access or exfiltration.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
87% confidence
Finding

The commands use the declared Canonry CLI, but perform an external submission action not covered by the stated analytical and review-oriented workflows. Sitemap submission is a distinct operational capability, not merely an implementation detail of inspecting evidence or reviewing integrations. The snippet contains example commands; it does not establish that any submission has actually occurred.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
86% confidence
Finding

The description emphasizes inspection, diagnosis, planning, review, and reporting, while the supplied commands also expose operational changes with persistent effects, potential costs, and outbound notifications. These are material capabilities beyond the stated observational and planning scope. The chunk is reference documentation, so it does not establish that any command runs automatically or that data is maliciously exfiltrated.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
85% confidence
Finding

Displaying a schedule fits the declared inspection and review purpose. However, schedule set --kind gbp-sync --preset daily changes operational configuration to schedule recurring synchronization, a state-changing capability not disclosed by the description's inspection, planning, review, and reporting scope.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
84% confidence
Finding

The non-dry-run canonry apply <file...> command introduces a change-applying capability that is not disclosed in the stated purpose. The dry-run variant supports inspection, but the regular apply operation goes beyond the described analytical workflows. The snippet does not establish which resources would be changed.

Content

No source excerpt is available for this finding.

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 262)May include surrounding context.

md
- [Indexing workflow after inspection and approval](references/indexing.md)

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 264)May include surrounding context.

md
- [Server-side traffic setup and diagnosis](references/server-side-traffic.md)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · references/canonry-cli.md (reported line 545)May include surrounding context.

md
uses `GET /projects/:name/content/dismissals`; dismiss uses `POST` with
`targetRef` and optional `addressedUrl` and `note`. Re-dismissing updates those
optional values and the dismissal time. Restore uses
`DELETE /projects/:name/content/dismissals/:targetRef` and returns HTTP 204; CLI and
MCP acknowledge it as `{ targetRef, restored: true }`. A missing record returns
the API's not-found error.

External Model or Provider Selection

High
Category
Excessive Agency
Confidence
90% confidence
Finding

Skill selects an external model or provider that may use a different account or billing plan than the operator expects. Undisclosed model switches can cause unexpected cost or quota consumption.

Content

Scanner excerpt · references/canonry-cli.md (reported line 730)May include surrounding context.

md
cnry settings                                  # show config: providers, apiUrl, db path
cnry settings --format json
cnry settings reload-providers --format json   # apply the server's saved provider config
cnry settings provider gemini --api-key <KEY> --model gemini-flash-latest
cnry settings provider openai --max-per-day 1000 --max-per-minute 20
cnry settings provider perplexity --api-key <KEY> --model fast
cnry settings provider muse --api-key <KEY> --model muse-spark-1.3

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · references/indexing.md (reported line 94)May include surrounding context.

cnry bing set-site https://example.com/

text

Get API key from: https://www.bing.com/webmasters/ → Settings → API Access

### Check connection and coverage
```bash

The skill directs agents to modify search-engine indexing state despite a manifest focused on inspection, diagnosis, planning, review, and reporting.

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

The reference directs sitemap submissions and bulk indexing requests as routine workflow steps without requiring approval for those external writes. Unlike simply documenting additional capabilities, directives such as 'submit immediately' can cause an agent responding to an inspection request to initiate crawler activity, consume quotas, or request expanded OAuth permissions without informed authorization.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/indexing.md (reported line 137)May include surrounding context.

Submit URLs:

bash
curl -X POST "https://www.bing.com/indexnow" \
  -H "Content-Type: application/json; charset=utf-8" \
  -d '{
    "host": "example.com",

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/server-side-traffic.md (reported line 407)May include surrounding context.

bash
CANONRY_PUBLIC_URL=https://canonry-ops.example.net

curl -sS -o /dev/null -w '%{http_code}\n' \
  -X POST -H 'content-type: application/json' --data '{}' \
  "${CANONRY_PUBLIC_URL%/}/api/v1/projects/<project>/traffic/cloudflare/ingest"

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/server-side-traffic.md (reported line 477)May include surrounding context.

bash
# assets -> no worker. Omitting "script" is what disables it.
curl -s -X POST -H "Authorization: Bearer $CF_TOKEN" -H "Content-Type: application/json" \
  --data '{"pattern":"<host>/_nuxt/*"}' \
  "https://api.cloudflare.com/client/v4/zones/$ZONE/workers/routes"