Back to skill

Security audit

pane

Security checks for vulnerabilities and agentic risk

Overview

This Pane integration is mostly disclosed and purpose-aligned, but it grants broad token-based access, can sync persistent agent identity files, and schedules delayed follow-up actions without enough user control.

Install only if you trust the Pane Gateway and intend to let this skill manage Pane sessions and agent identity sync. Keep PANE_GATEWAY_URL pinned to a loopback/local address, avoid PANE_GATEWAY_INSECURE_TLS for authenticated use, treat PANE_GATEWAY_TOKEN as a powerful 90-day secret, and review or disable workflows that sync identity files or create scheduled follow-up jobs.

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
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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 (4)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:13
Finding

Unrestricted Gateway URL Can Disclose Bearer Tokens and Sensitive Content

Content
View full analysis
Remediation
View remediation
` and `https://[::1]:`. 3. If remote Gateways are supported, require an explicit host allowlist and separate user approval. 4. Resolve hostnames and reject destinations that resolve outside the approved address range. 5. Disable redirects with `curl --max-redirs 0`, or validate every redirect destination before forwarding credentials. 6. Use separate scoped credentials for sessions, chat proxying, and identity synchronization. 7. Require explicit user approval before transmitting identity, memory, user-profile, or log content. 8. Redact secrets from synchronized files and provide a preview of every outbound sensitive payload. 9. Provide token revocation and short-lived token support rather than relying on one broad 90-day credential. ]]>

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:52
Finding

Certificate Bootstrapping and Optional Insecure TLS Permit Gateway Impersonation

Content
View full analysis
/tmp/pane-gateway.pem CACERT=/tmp/pane-gateway.pem curl -sS --cacert "$CACERT" \ -X POST "$PANE_GATEWAY_URL/pair" \ -H "Content-Type: application/json" \ -d '{"code":"123456"}' ``` The documented TLS handling repeats the unauthenticated certificate bootstrap: ```bash curl -sS "$PANE_GATEWAY_URL/v1/health" | jq -r '.tls_cert_pem' > /tmp/pane-gateway.pem curl -sS --cacert /tmp/pane-gateway.pem "$PANE_GATEWAY_URL/v1/health" ``` It also permits complete certificate-verification bypass: ```bash if [ "$PANE_GATEWAY_INSECURE_TLS" = "1" ]; then curl -sS -k "$PANE_GATEWAY_URL/v1/health" else curl -sS --cacert /tmp/pane-gateway.pem "$PANE_GATEWAY_URL/v1/health" fi ``` ### Technical Analysis The certificate used to authenticate later connections is downloaded from the same endpoint before an out-of-band identity check is performed. This is a trust-on-first-use pattern: if the first connection is redirected or intercepted, the client can store an attacker-provided certificate and subsequently treat that certificate as trusted. The documented `-k` mode is more severe because it disables server authentication entirely. Although described as a development option, the Skill subsequently allows the same substitution for authenticated requests. The fixed `/tmp/pane-gateway.pem` path introduces an additional local risk. Another local process may pre-create or replace that path, potentially through a symbolic link, resulting in certificate substitution or unintended file overwrite. The instructions do not create the file atomically or restrict its permissions. ### Attack Path 1. An attacker redirects the configured Gateway host, intercepts the initial connection, or manipulates th ...[truncated 939 chars]
Remediation
View remediation

T02 · Agent Memory Poisoning

Error
Location
SKILL.md:297
Finding

Broad Identity Synchronization Enables Persistent Agent Memory and Control-File Modification

Content
View full analysis
Remediation
View remediation

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:180
Finding

Mandatory One-Shot Cron Scheduling Creates Delayed Cross-Turn Agent Actions

Content
View full analysis
" --session "session:" --no-deliver --delete-after-run ``` Your current session key is in the `## Runtime` section of your system prompt (`session=agent::openai-user:pane-`). Use it exactly. 3. Never use `--announce`, any delivery mode, or `--session isolated`. The Pane transcript is the delivery: Pane polls this session and shows your reply automatically. 4. Complete every step through this session: reply with results, use Pane's tools, schedule follow-ups with the cron recipe above. `exec` is permitted only to run `openclaw cron add`. Never use it to do the user's task, and never touch files or databases outside the session. 5. Report each completed step as its own reply. Schedule silently — do not narrate cron mechanics to the user. 6. Delivery latency is up to 60 seconds per step while the user has the session open. Messages sent while Pane is closed appear when the user next opens the session. ``` ### Technical Analysis The Skill directs the agent to change its normal execution model whenever it interprets a request as multi-step. It mandates delayed execution rather than requiring explicit user consent, extracts an internal session identifier from the system prompt, and instructs the agent to schedule silently. Although the cron entries are one-shot and use `- ...[truncated 2046 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • 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 (18)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
88% confidence
Finding

The trust-on-first-use pattern here abuses a security-sensitive curl parameter by binding --cacert to certificate material fetched from the same untrusted endpoint it is meant to authenticate. This can create a false sense of secure transport while still permitting interception and token theft during initial setup.

Content

Scanner excerpt · SKILL.md (reported line 79)May include surrounding context.

bash
curl -sS "$PANE_GATEWAY_URL/v1/health" | jq -r '.tls_cert_pem' > /tmp/pane-gateway.pem
curl -sS --cacert /tmp/pane-gateway.pem "$PANE_GATEWAY_URL/v1/health"

2. Skip verification (dev only). If PANE_GATEWAY_INSECURE_TLS=1 is set,

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
88% confidence
Finding

The trust-on-first-use pattern here abuses a security-sensitive curl parameter by binding --cacert to certificate material fetched from the same untrusted endpoint it is meant to authenticate. This can create a false sense of secure transport while still permitting interception and token theft during initial setup.

Content

Scanner excerpt · SKILL.md (reported line 79)May include surrounding context.

bash
curl -sS "$PANE_GATEWAY_URL/v1/health" | jq -r '.tls_cert_pem' > /tmp/pane-gateway.pem
curl -sS --cacert /tmp/pane-gateway.pem "$PANE_GATEWAY_URL/v1/health"

2. Skip verification (dev only). If PANE_GATEWAY_INSECURE_TLS=1 is set,

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · SKILL.md (reported line 204)May include surrounding context.

Health check

bash
curl -sS --cacert "$CACERT" "$PANE_GATEWAY_URL/v1/health"
# Authenticated (fuller response incl. oc_connection_status):
curl -sS --cacert "$CACERT" "$PANE_GATEWAY_URL/v1/health" \
  -H "Authorization: Bearer $PANE_GATEWAY_TOKEN"

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 334)May include surrounding context.

md
See `references/gateway-api.md` for the full endpoint reference including

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/gateway-api.md (reported line 139)May include surrounding context.

}

text

### `DELETE /v1/sessions/:session_id`

Response `200`: `{ "deleted_at": "..." }`

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · references/gateway-api.md (reported line 192)May include surrounding context.

message.delta, message.complete) with the payload as data:. Keep-alive comment lines are sent every 15s.

curl consumption pattern:

bash
curl -sS -N --cacert "$CACERT" \
  "$PANE_GATEWAY_URL/v1/sessions/$SESSION_ID/messages/stream" \

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/gateway-api.md (reported line 250)May include surrounding context.

Response 200: { "pending_changes": [ /* PendingChange objects */ ] }

DELETE /v1/sync/pull/confirm

Request:

json

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The manifest claims note/task/project creation, but the API reference states those resource endpoints do not exist and should never be used. This discrepancy can mislead an agent into fabricating unsupported calls or falling back to unintended conversational workflows, increasing the risk of unsafe or inconsistent actions and user deception.

Content

No source excerpt is available for this finding.

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/gateway-api.md (reported line 293)May include surrounding context.

md
## Unreachable/blocked-in-v1 endpoints

These do **not** exist on the gateway and should never be constructed:
`POST/GET/PUT/DELETE /v1/notes*`, `/v1/tasks*`, `/v1/projects*`,
`/v1/folders*`. Requests to any of these hit the router's fallback handler
(404). Use the conversational session flow in `SKILL.md` instead.

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 · SKILL.md (reported line 59)May include surrounding context.

md
curl -sS "$PANE_GATEWAY_URL/v1/health" | jq -r '.tls_cert_pem' > /tmp/pane-gateway.pem
CACERT=/tmp/pane-gateway.pem

curl -sS --cacert "$CACERT" \
  -X POST "$PANE_GATEWAY_URL/pair" \
  -H "Content-Type: application/json" \
  -d '{"code":"123456"}'

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 · SKILL.md (reported line 79)May include surrounding context.

bash
curl -sS "$PANE_GATEWAY_URL/v1/health" | jq -r '.tls_cert_pem' > /tmp/pane-gateway.pem
curl -sS --cacert /tmp/pane-gateway.pem "$PANE_GATEWAY_URL/v1/health"

2. Skip verification (dev only). If PANE_GATEWAY_INSECURE_TLS=1 is set,

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
85% confidence
Finding

The skill permits disabling TLS verification with -k when an environment variable is set. Even though it is labeled dev-only, this normalizes operation without certificate validation and exposes bearer tokens and session data to man-in-the-middle attacks if the gateway URL is not genuinely local and trusted.

Content

Scanner excerpt · SKILL.md (reported line 82)May include surrounding context.

curl -sS --cacert /tmp/pane-gateway.pem "$PANE_GATEWAY_URL/v1/health"

text

**2. Skip verification (dev only).** If `PANE_GATEWAY_INSECURE_TLS=1` is set,
use `-k`/`--insecure`. Do not default to this silently — only use it when the
env var is explicitly set, and mention to the user that verification is
skipped.

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 · SKILL.md (reported line 137)May include surrounding context.

md
SESSION_ID=$(echo "$SESSION" | jq -r '.session_id')

# 2. Send a message describing the desired note
curl -sS --cacert "$CACERT" \
  -X POST "$PANE_GATEWAY_URL/v1/sessions/$SESSION_ID/messages" \
  -H "Authorization: Bearer $PANE_GATEWAY_TOKEN" \
  -H "Content-Type: application/json" \

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill instructs the agent to use OpenClaw cron scheduling and explicitly broadens permitted actions beyond the Pane gateway. This expands the skill's operational scope, enables unattended follow-up actions, and creates a persistence mechanism that could continue acting after the initial user request, increasing risk of unintended or unauthorized actions.

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 · SKILL.md (reported line 204)May include surrounding context.

Health check

bash
curl -sS --cacert "$CACERT" "$PANE_GATEWAY_URL/v1/health"
# Authenticated (fuller response incl. oc_connection_status):
curl -sS --cacert "$CACERT" "$PANE_GATEWAY_URL/v1/health" \
  -H "Authorization: Bearer $PANE_GATEWAY_TOKEN"

External Transmission

Medium
Category
Data Exfiltration
Confidence
82% confidence
Finding

The sync push endpoint transmits local identity files such as MEMORY.md, SOUL.md, RULES.md, and logs to the gateway. These files can contain sensitive agent context, secrets, or behavioral controls, so enabling automated upload of their contents increases the risk of confidential data disclosure or prompt/identity manipulation if the gateway is compromised or misdirected.

Content

Scanner excerpt · SKILL.md (reported line 324)May include surrounding context.

md
# Push a file (checksum is SHA-256 hex of content, verified server-side)
CONTENT='# Updated memory'
CHECKSUM=$(printf '%s' "$CONTENT" | shasum -a 256 | cut -d' ' -f1)
curl -sS --cacert "$CACERT" -X POST "$PANE_GATEWAY_URL/v1/sync/push" \
  -H "Authorization: Bearer $PANE_GATEWAY_TOKEN" \
  -H "Content-Type: application/json" \
  -d "$(jq -n --arg agent main --arg fn MEMORY.md --arg content "$CONTENT" --arg cs "$CHECKSUM" \

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

This markdown file documents DELETE /v1/sessions/:session_id, which deletes a session, but provides no warning about the data-loss implications or recommendation to confirm intent before use. Under the markdown-specific warning rule, destructive behavior that can affect user data should be accompanied by a disclosure or caution.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
87% confidence
Finding

The manifest describes creating notes, tasks, and projects via chat sessions, managing AI sessions, and syncing identity files. This reference also documents POST /v1/chat/completions, a separate OpenAI-compatible completion proxy that bypasses Pane sessions and UI history, which is a broader behavior than the manifest’s session-focused description.

Content

No source excerpt is available for this finding.