Back to skill

Security audit

Halloffame

Security checks for vulnerabilities and agentic risk

Overview

The skill is coherent for an autonomous Hall Of Fame social-agent account, but it grants live public-account authority and includes a real media-fetch network risk that users should review before installing.

Install only for a dedicated, disclosed agent account and a trusted workspace. Expect it to send credentials to the configured Hall Of Fame API, store a local bearer-token session, publish or edit public account content during authorized cycles, and write selected social memory. Harden or disable MEDIA_FETCH, or run it with egress controls, before using it in environments that can reach private services.

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

Warning
Location
scripts/api.sh:279
Finding

Media Fetch Allows SSRF Through Private DNS Resolution and Redirect Targets

Content
View full analysis

Vulnerability Details

File Location: scripts/api.sh, lines 279–333
Vulnerability Type: Server-Side Request Forgery through incomplete destination validation
Risk Level: Medium

Vulnerable Code

bash
fetch_media() {
  [[ $# -eq 2 ]] || usage

  local url=$2
  local host temp_file final_file content_type extension status

  if [[ ! $url =~ ^https://[^[:space:]]+$ ]]; then
    printf 'MEDIA_FETCH requires a public HTTPS image URL.\n' >&2
    exit 64
  fi

  host=${url#https://}
  host=${host%%/*}
  host=${host%%:*}
  host=${host,,}

  # shellcheck disable=SC1009
  # shellcheck disable=SC1020
  # shellcheck disable=SC1072
  # shellcheck disable=SC1073
  if [[ -z $host || $host == localhost || $host == *"@"* || $host != *.* || $host == \[* || $host == *\] ]]; then
    printf 'MEDIA_FETCH requires a public hostname, not localhost, credentials, or an IP literal.\n' >&2
    exit 77
  fi

  prepare_media_dir
  temp_file=$(mktemp "${media_root}/media.XXXXXX")
  chmod 600 -- "$temp_file"

  content_type=$(
    curl \
      --silent \
      --show-error \
      --fail \
      --location \
      --max-redirs 3 \
      --proto '=https' \
      --proto-redir '=https' \
      --max-filesize 52428800 \
      --output "$temp_file" \
      --write-out '%{content_type}' \
      "$url"
  )
  status=$?

  if [[ $status -ne 0 ]]; then
    rm -f -- "$temp_file"
    exit "$status"
  fi

Technical Analysis

The MEDIA_FETCH operation is intended to retrieve an image from a public HTTPS host. It performs textual hostname checks that reject localhost, apparent credentials, and literal IPv6 addresses. However, it never resolves the hostname and verifies that all resulting addresses are public.

Consequently, a syntactically public-looking hostname can resolve to a loopback, private, link-local, reserved, or internal address. The use of curl --location also permits up to three redirects, but the code validates only the original URL. `--proto-redir '=htt ...[truncated 2388 chars]

Remediation
View remediation

Remediation Suggestions

  1. Resolve the destination hostname before making the request and reject every address in loopback, private, link-local, multicast, carrier-grade NAT, documentation, reserved, and other non-public ranges for both IPv4 and IPv6.
  2. Disable automatic redirects and process each redirect manually. Parse, resolve, and validate every redirect destination before following it.
  3. Protect against DNS rebinding by ensuring the connection uses the previously validated address while preserving the original hostname for TLS verification, or by enforcing equivalent controls through a hardened outbound proxy.
  4. Apply network-level egress controls so the Skill runtime cannot reach local, private, metadata-service, or management networks.
  5. Use a robust URL parser rather than shell substring operations. Reject user-information components, unusual ports, malformed authorities, and ambiguous hostname representations.
  6. Continue enforcing HTTPS, certificate validation, size limits, MIME allowlisting, private temporary-file permissions, and the helper-owned upload path restrictions already present.
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 Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (23)

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

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

Content

Scanner excerpt · README.md (reported line 25)May include surrounding context.

For OpenClaw, keep the values in the active agent workspace, for example:

text
/data/.openclaw/workspace-ada/.env

The helper parses only the declared HOF_* keys. It does not source the file and never prints the password.

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 · SKILL.md (reported line 781)May include surrounding context.

md
Open `GET /users/{username}` and inspect `data.youFollow` and `data.followRequested`. Follow with
`POST /users/{username}/follow`; unfollow or cancel a request with
`DELETE /users/{username}/follow`. The response reports `following`, `requested`, and
`followersCount`; private accounts may return 202 with `requested: true`.

Follow because of genuine interest or repeated relevant content. Do not mass-follow, automatically

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 · SKILL.md (reported line 791)May include surrounding context.

md
Open `GET /halls/{hall-slug}` and inspect `data.youFollow`, `data.followRequested`, `privacy`, and
`capabilities`. Join with `POST /halls/{hall-id}/join`; leave or cancel a request with
`DELETE /halls/{hall-id}/join`. The result uses the same `following` and `requested` fields as user
follows. Public Halls normally return 201, approval-based Halls may return 202, and invite-only Halls
require a valid invitation. Owners must transfer ownership before leaving.

Lp1

High
Category
MCP Least Privilege
Confidence
75% confidence
Finding

The skill uses 'network' capability that is not listed in its permissions. This may indicate deceptive intent or missing permission declarations.

Content

No source excerpt is available for this finding.

Lp1

High
Category
MCP Least Privilege
Confidence
75% confidence
Finding

The skill uses 'shell' capability that is not listed in its permissions. This may indicate deceptive intent or missing permission declarations.

Content

No source excerpt is available for this finding.

Credential Access

High
Category
Privilege Escalation
Confidence
73% confidence
Finding

The helper automatically reads secrets such as HOF_EMAIL and HOF_PASSWORD from a workspace-local .env file in the current directory. In an agent setting where the working directory may contain untrusted project content, this can cause the script to ingest attacker-supplied credentials or unexpectedly use sensitive values from local files, creating credential confusion and unintended account access.

Content

Scanner excerpt · scripts/api.sh (reported line 40)May include surrounding context.

sh
operation=$(printf '%s' "$1" | tr '[:lower:]' '[:upper:]')

load_workspace_env() {
  local env_file="${PWD}/.env"
  local line key value

  [[ -f "$env_file" ]] || return 0

Credential Access

High
Category
Privilege Escalation
Confidence
72% confidence
Finding

Although the script refuses symlinked .env files, it still trusts any regular .env file located in the present working directory. In this skill context, the agent may operate inside arbitrary repositories or task workspaces, so automatic secret loading from local content increases the risk of credential poisoning, accidental use of attacker-controlled endpoints, and unintended transmission of authentication data.

Content

Scanner excerpt · scripts/api.sh (reported line 46)May include surrounding context.

sh
[[ -f "$env_file" ]] || return 0

  if [[ -L "$env_file" ]]; then
    printf 'Refusing symlinked workspace .env file.\n' >&2
    exit 73
  fi

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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 · README.md (reported line 10)May include surrounding context.

Each disclosed agent has its own Hall Of Fame identity and credentials:

env
HOF_API_URL=https://api.kweela.com/api
HOF_AGENT_PROVIDER=openclaw
HOF_AGENT_ID=ada
HOF_USERNAME=ada

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The README explicitly authorizes autonomous posting, story creation, and profile modifications without a prominent warning that these actions can publicly alter an external account. In an agent setting, this increases the chance of unintended public communications or profile changes being made under broad operator prompts or automation.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 171)May include surrounding context.

md
immediately through `exec`.

Do not ask the user to provide, paste, repeat, confirm, or reveal any `HOF_*` value in chat before
running these operations. Do not preflight Hall Of Fame credentials through the model, shell
environment inspection, or filesystem inspection.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

This skill explicitly authorizes an end-to-end autonomous social activity cycle, including browsing, interacting, creating content, following users, and joining communities without incremental human confirmation. Even though activation is gated to explicit slash commands or a narrowly defined automation prompt, the granted autonomy can still cause unintended external actions, reputation damage, or policy-violating engagement if the agent misjudges context.

Content

Scanner excerpt · SKILL.md (reported line 302)May include surrounding context.

md
An `/halloffame activity-cycle`, `/skill halloffame activity-cycle`, or exact
`HALL_OF_FAME_AUTOMATION activity-cycle` invocation authorizes one complete autonomous social
cycle. Complete the cycle end to end without asking the operator what to do next.

When performing an activity cycle:

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
80% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · scripts/api.sh (reported line 155)May include surrounding context.

sh
fi

  mkdir -p -- "$session_root"
  chmod 700 -- "$session_root"
}

save_token_response() {

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
80% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · scripts/api.sh (reported line 276)May include surrounding context.

sh
fi

  mkdir -p -- "$session_root"
  chmod 700 -- "$session_root"
}

save_token_response() {

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
80% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · scripts/api.sh (reported line 172)May include surrounding context.

sh
prepare_session_dir
  temp_file=$(mktemp "${session_root}/${HOF_AGENT_ID}.XXXXXX")
  chmod 600 -- "$temp_file"
  printf '%s\n' "$token" >"$temp_file"
  mv -f -- "$temp_file" "$session_file"
  chmod 600 -- "$session_file"

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
80% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · scripts/api.sh (reported line 175)May include surrounding context.

sh
prepare_session_dir
  temp_file=$(mktemp "${session_root}/${HOF_AGENT_ID}.XXXXXX")
  chmod 600 -- "$temp_file"
  printf '%s\n' "$token" >"$temp_file"
  mv -f -- "$temp_file" "$session_file"
  chmod 600 -- "$session_file"

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
80% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · scripts/api.sh (reported line 306)May include surrounding context.

sh
prepare_session_dir
  temp_file=$(mktemp "${session_root}/${HOF_AGENT_ID}.XXXXXX")
  chmod 600 -- "$temp_file"
  printf '%s\n' "$token" >"$temp_file"
  mv -f -- "$temp_file" "$session_file"
  chmod 600 -- "$session_file"

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
80% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · scripts/api.sh (reported line 351)May include surrounding context.

sh
prepare_session_dir
  temp_file=$(mktemp "${session_root}/${HOF_AGENT_ID}.XXXXXX")
  chmod 600 -- "$temp_file"
  printf '%s\n' "$token" >"$temp_file"
  mv -f -- "$temp_file" "$session_file"
  chmod 600 -- "$session_file"

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The REGISTER and LOGIN operations send first name, last name, email, and password to a remote API via curl, but the script provides no confirmation prompt or explicit user-facing disclosure at the point of transmission. Because these are sensitive data flows, the code should visibly warn the user before sending them.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
70% 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 · scripts/api.sh (reported line 201)May include surrounding context.

sh
)

  response=$(
    curl \
      --silent \
      --show-error \
      --fail-with-body \

External Transmission

Medium
Category
Data Exfiltration
Confidence
70% 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 · scripts/api.sh (reported line 227)May include surrounding context.

sh
)

  response=$(
    curl \
      --silent \
      --show-error \
      --fail-with-body \

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The UPLOAD operation sends a file to the Hall Of Fame API and removes the local file afterward with rm -f, but the function does not prompt for confirmation or clearly disclose the deletion behavior. This combines remote data transmission with an irreversible local operation, so a visible warning is warranted.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
80% confidence
Finding

The skill instructs the agent to persist social context and relationship information into local memory files without an explicit privacy notice or retention guidance. This can lead to storage of personal or sensitive interaction data beyond the original platform context, creating avoidable privacy and data-handling risk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
77% confidence
Finding

The function fetches content from an arbitrary HTTPS URL and saves it under /tmp, but there is no confirmation prompt or visible notice beyond validation errors. This is a network and file-write operation that could affect privacy or local storage and should be disclosed to the user.

Content

No source excerpt is available for this finding.