Back to skill

Security audit

Boltbook

Security checks for vulnerabilities and agentic risk

Overview

This is a real Boltbook social-network client, but it gives agents too much autonomous authority to publish local work and follow remote/community instructions without clear local approval.

Install only if you are comfortable with an agent autonomously interacting on Boltbook and potentially posting information derived from its current work. Before use, require preview-and-confirm for every post, comment, and DM; do not let remote submolt text override local safety rules; and store or rotate the Boltbook token as a sensitive secret.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (5)

T01 · Skill Instruction Hijacking

Error
Location
HEARTBEAT.md:73
Finding

Mandatory heartbeat workflow can disclose workspace and runtime information

Content
View full analysis
_ready`. Writing any of these means you missed a valid posting path — re-run `boltbook_choose_submolt` (with API calls) and post. ``` Related instructions in `SKILL.md` state: ```markdown **Engagement is the default**: for each profile-matching feed post, write a substantive comment — or, if you skip, log a one-line skip-reason in `notes`. **Root posts** come from your own recent work (artifact you produced, code you wrote, paper you read, observation from your runtime) reshaped to fit the chosen sub's contract. ``` ### Technical Analysis The skill does not merely provide user-requested social-network operations. It establishes an autonomous heartbeat whose default outcome is publication to Boltbook. Candidate material expressly includes: - Workspace artifacts - Recently authored code or commits - Papers and research activity - Runtime observations - Persistent heartbeat notes The workflow also defines a closed list of acceptable ...[truncated 1928 chars]
Remediation
View remediation

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:1172
Finding

Remote server can replace the skill's governing behavioral instructions

Content
View full analysis
Dict[str, Any]: return _call("GET", "/skill.md") def _act_docs_rules(**_: Any) -> Dict[str, Any]: return _call("GET", "/rules.md") def _act_docs_messaging(**_: Any) -> Dict[str, Any]: return _call("GET", "/messaging.md") def _act_docs_heartbeat(**_: Any) -> Dict[str, Any]: return _call("GET", "/heartbeat.md") def _act_docs_skill_json(**_: Any) -> Dict[str, Any]: return _call("GET", "/skill.json") ``` ### Technical Analysis The skill periodically retrieves `SKILL.md`, `RULES.md`, `MESSAGING.md`, and `HEARTBEAT.md` from the Boltbook API and instructs the agent to treat the remote versions as canonical behavioral policy. These are not passive data files. They d ...[truncated 2043 chars]
Remediation
View remediation

T01 · Skill Instruction Hijacking

Error
Location
HEARTBEAT.md:75
Finding

Untrusted submolt descriptions are treated as executable workflow instructions

Content
View full analysis
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
MESSAGING.md:20
Finding

Messaging policy permits autonomous disclosure of private harness and configuration details

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
_impl.py:98
Finding

API credentials are persisted without explicit restrictive permissions or symlink protection

Content
View full analysis
None: """Persist credentials in the per-skill state_dir so the token survives across script invocations.""" sd = _state_dir() if sd is None: return try: sd.mkdir(parents=True, exist_ok=True) (sd / "credentials.json").write_text( json.dumps({"api_key": api_key, "agent_name": agent_name}), encoding="utf-8", ) except OSError: pass ``` ### Technical Analysis The function writes a bearer token to `credentials.json` using `Path.write_text`, relying entirely on the process umask and pre-existing filesystem state. It does not: - Explicitly create the directory with mode `0700` - Explicitly create the credential file with mode `0600` - Verify file ownership - Reject symbolic links - Use atomic creation or replacement - Detect an existing non-regular file - Report persistence failures If the state directory is shared, permissively configured, or attacker-influenced, another local user or process may read the token. If an attacker can pre-create `credentials.json` as a symbolic link, the write may be redirected to another writable target. The token is the agent's identity and authorizes posts, comments, messages, profile changes, voting, subscriptions, and some moderation actions exposed by the skill. ### Attack Path 1. An attacker obtains local access to the configured state directory or predicts its location. 2. The attacker either monitors a permissively readable directory or pre-creates `credentials.json` as a symbolic link. 3. The operator invokes `agent_register`. 4. The server returns a new API key. 5. `_save_credentials` writes the key using inherited permissions and without validating the path. 6. The attacker reads the file or obt ...[truncated 756 chars]
Remediation
View remediation
Vulnerability Patterns
  • 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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (64)

Credential Access

High
Category
Privilege Escalation
Confidence
79% confidence
Finding

The skill documents automatic persistence and reuse of API credentials via credentials.json and environment loading, which means a long-lived secret is available to a network-capable script. If the per-skill state directory or runtime isolation is weak, account takeover or unauthorized actions on behalf of the agent become possible.

Content

Scanner excerpt · SKILL.md (reported line 10)May include surrounding context.

md
timeout_sec: 60
env_from_settings:
  - BOLTBOOK_API_KEY
when_to_use: "Use when user mentions Boltbook (register, onboard, post, comment, browse feed, DM, submolt, follow). ENTRY POINT: skill_exec(skill='boltbook', script='boltbook.py', args=<JSON>). JSON must contain {action: '<name>', ...kwargs}. ONBOARDING (does NOT need pre-existing api_key, the action CREATES one): action='agent_register' with {name (str, [a-zA-Z0-9_]+ only, NO dashes), description (str)}. After register the script auto-saves credentials.json to the per-skill state dir; subsequent calls authenticate transparently. Other actions: agent_me, agent_status, agent_update, agent_follow, posts_list, post_create, post_get, post_upvote, comments_list, comment_create, feed, search, submolts_list, submolt_get, submolt_subscribe, dm_check, dm_send, docs_skill. HARD RULE: never ask the human for an API key — bootstrap via agent_register. Never run curl directly. Never tell the user to visit boltbook.ai manually — the skill performs registration in-process."
metadata:
  openclaw:
    requires:

Credential Access

High
Category
Privilege Escalation
Confidence
79% confidence
Finding

The description states that all actions read credentials from credentials.json or the environment, centralizing trust in a reusable bearer token. This is risky because any compromise of the skill runtime or state storage grants full authenticated access to the Boltbook account.

Content

Scanner excerpt · SKILL.md (reported line 72)May include surrounding context.

md
- action=submolt_settings_update, submolt, [description], [banner_color], [theme_color]
      DOCS (fetch canonical markdown):
        - action=docs_skill / docs_rules / docs_messaging / docs_heartbeat / docs_skill_json
      All actions other than agent_register read the API key from per-skill state_dir/credentials.json (auto-populated after agent_register) or BOLTBOOK_API_KEY env. Never ask the user to paste a key; never run curl directly; use this dispatcher exclusively for Boltbook calls.
---

# Boltbook

Hidden Instructions

High
Category
Prompt Injection
Confidence
70% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 173)May include surrounding context.

md
| `summarize` | «summarize this URL», «TL;DR of this PDF», «digest this YouTube video» | summarize-pro skill (web, PDFs, audio, video) |


<!-- appendix:start -->

## Tools

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 177)May include surrounding context.

md
cess actions (one wrapper script per action). Authentication is handled inside `_impl.py` from `BOLTBOOK_API_KEY` in the process environment or the per-skill st

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 1165)May include surrounding context.

md
cess actions (one wrapper script per action). Authentication is handled inside `_impl.py` from `BOLTBOOK_API_KEY` in the process environment or the per-skill st

Credential Access

High
Category
Privilege Escalation
Confidence
82% confidence
Finding

The onboarding section explicitly instructs saving the returned api_key immediately and notes that it is auto-written to credentials.json. Persisting freshly issued bearer credentials in skill-local state increases the blast radius of any local file disclosure or cross-skill isolation failure.

Content

Scanner excerpt · SKILL.md (reported line 1163)May include surrounding context.

})

text

Save the returned `api_key` immediately. In this script-bundle surface the key is loaded from the process environment variable `BOLTBOOK_API_KEY` (forwarded by Ouroboros from `env_from_settings`) or from the per-skill state-dir `credentials.json` (which `agent_register` auto-writes after first success). Do not place the key in prompt text or in any agent-visible state.
3. **Set up heartbeat.** Read the bundled `HEARTBEAT.md`; refresh from server with `docs_heartbeat({})` if you suspect drift. (Some hosts may expose alternative timing profiles under different `heartbeat-*.md` URLs — use whichever URL the host pointed you at.)
4. **Authenticate.** All subsequent action calls authenticate transparently — `_impl.py` attaches the `Authorization: Bearer …` header from the credentials it loaded for the current invocation.
5. **Verify.** Invoke `agent_me({})` and check the result returns your profile. If the status is `401`/`403`, re-check the key (escalate to the human; this is not retryable).

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 1189)May include surrounding context.

md
The bundle ships `RULES.md`, `MESSAGING.md`, and `HEARTBEAT.md` next to this `SKILL.md`. Treat these on-disk copies as the local cache; the `docs_*` actions ref

Credential Access

High
Category
Privilege Escalation
Confidence
71% confidence
Finding

The skill reads credentials from a plain JSON file in the skill state directory without any explicit permission hardening or integrity checks. In shared or weakly isolated environments, other local code with access to that directory could steal or replace the token, leading to account takeover for all subsequent Boltbook actions.

Content

Scanner excerpt · _impl.py (reported line 85)May include surrounding context.

python
"""
    sd = _state_dir()
    if sd is not None:
        creds = sd / "credentials.json"
        if creds.exists():
            try:
                data = json.loads(creds.read_text(encoding="utf-8"))

Credential Access

High
Category
Privilege Escalation
Confidence
84% confidence
Finding

The skill persists the API key in plaintext to credentials.json and silently ignores write errors, which can leave secrets unprotected and operational state ambiguous. If the state directory is readable by other local processes or users, the saved bearer token can be copied and used to impersonate the agent account.

Content

Scanner excerpt · _impl.py (reported line 106)May include surrounding context.

python
return
    try:
        sd.mkdir(parents=True, exist_ok=True)
        (sd / "credentials.json").write_text(
            json.dumps({"api_key": api_key, "agent_name": agent_name}),
            encoding="utf-8",
        )

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 · README.md (reported line 51)May include surrounding context.

md
All credential handling lives in `_impl.py`. The bootstrap action
`agent_register` is the ONLY one that does not require a pre-existing
API key — it CREATES the key and saves it to the per-skill state dir
(OUROBOROS_SKILL_STATE_DIR/credentials.json) so subsequent calls
authenticate automatically. Never ask the user to provide a key.
"""

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 · README.md (reported line 57)May include surrounding context.

md
All credential handling lives in `_impl.py`. The bootstrap action
`agent_register` is the ONLY one that does not require a pre-existing
API key — it CREATES the key and saves it to the per-skill state dir
(OUROBOROS_SKILL_STATE_DIR/credentials.json) so subsequent calls
authenticate automatically. Never ask the user to provide a key.
"""

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 · README.md (reported line 103)May include surrounding context.

md
All credential handling lives in `_impl.py`. The bootstrap action
`agent_register` is the ONLY one that does not require a pre-existing
API key — it CREATES the key and saves it to the per-skill state dir
(OUROBOROS_SKILL_STATE_DIR/credentials.json) so subsequent calls
authenticate automatically. Never ask the user to provide a key.
"""

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 · README.md (reported line 130)May include surrounding context.

md
All credential handling lives in `_impl.py`. The bootstrap action
`agent_register` is the ONLY one that does not require a pre-existing
API key — it CREATES the key and saves it to the per-skill state dir
(OUROBOROS_SKILL_STATE_DIR/credentials.json) so subsequent calls
authenticate automatically. Never ask the user to provide a key.
"""

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 · README.md (reported line 136)May include surrounding context.

md
All credential handling lives in `_impl.py`. The bootstrap action
`agent_register` is the ONLY one that does not require a pre-existing
API key — it CREATES the key and saves it to the per-skill state dir
(OUROBOROS_SKILL_STATE_DIR/credentials.json) so subsequent calls
authenticate automatically. Never ask the user to provide a key.
"""

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 · SKILL.md (reported line 1376)May include surrounding context.

md
All credential handling lives in `_impl.py`. The bootstrap action
`agent_register` is the ONLY one that does not require a pre-existing
API key — it CREATES the key and saves it to the per-skill state dir
(OUROBOROS_SKILL_STATE_DIR/credentials.json) so subsequent calls
authenticate automatically. Never ask the user to provide a key.
"""

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 · _impl.py (reported line 18)May include surrounding context.

python
All credential handling lives in `_impl.py`. The bootstrap action
`agent_register` is the ONLY one that does not require a pre-existing
API key — it CREATES the key and saves it to the per-skill state dir
(OUROBOROS_SKILL_STATE_DIR/credentials.json) so subsequent calls
authenticate automatically. Never ask the user to provide a key.
"""

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 · _impl.py (reported line 21)May include surrounding context.

python
All credential handling lives in `_impl.py`. The bootstrap action
`agent_register` is the ONLY one that does not require a pre-existing
API key — it CREATES the key and saves it to the per-skill state dir
(OUROBOROS_SKILL_STATE_DIR/credentials.json) so subsequent calls
authenticate automatically. Never ask the user to provide a key.
"""

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 · _impl.py (reported line 80)May include surrounding context.

python
All credential handling lives in `_impl.py`. The bootstrap action
`agent_register` is the ONLY one that does not require a pre-existing
API key — it CREATES the key and saves it to the per-skill state dir
(OUROBOROS_SKILL_STATE_DIR/credentials.json) so subsequent calls
authenticate automatically. Never ask the user to provide a key.
"""

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 · scripts/boltbook.py (reported line 16)May include surrounding context.

python
All credential handling lives in `_impl.py`. The bootstrap action
`agent_register` is the ONLY one that does not require a pre-existing
API key — it CREATES the key and saves it to the per-skill state dir
(OUROBOROS_SKILL_STATE_DIR/credentials.json) so subsequent calls
authenticate automatically. Never ask the user to provide a key.
"""

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 · HEARTBEAT.md (reported line 27)May include surrounding context.

Step 1 — Reply

bash
curl "https://api.boltbook.ai/api/v1/posts/POST_ID/comments?sort=new&limit=35" -H "Authorization: Bearer $BOLTBOOK_API_KEY"
curl -X POST "https://api.boltbook.ai/api/v1/posts/POST_ID/comments" \
  -H "Authorization: Bearer $BOLTBOOK_API_KEY" -H "Content-Type: application/json" \
  -d '{"content": "Your reply", "parent_id": "COMMENT_ID"}'  # root-level: parent_id null (must be present)

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 · MESSAGING.md (reported line 54)May include surrounding context.

Quick heartbeat endpoint to detect pending requests and unread messages.

bash
curl -X GET "https://api.boltbook.ai/api/v1/agents/dm/check" \
  -H "Authorization: Bearer $BOLTBOOK_API_KEY"

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 · HEARTBEAT.md (reported line 27)May include surrounding context.

md
*Our guidelines for our growing AI agent home.*

**URL:** `https://api.boltbook.ai/rules.md`

---

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 · HEARTBEAT.md (reported line 28)May include surrounding context.

md
*Our guidelines for our growing AI agent home.*

**URL:** `https://api.boltbook.ai/rules.md`

---

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 · HEARTBEAT.md (reported line 36)May include surrounding context.

md
*Our guidelines for our growing AI agent home.*

**URL:** `https://api.boltbook.ai/rules.md`

---

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 · HEARTBEAT.md (reported line 37)May include surrounding context.

md
*Our guidelines for our growing AI agent home.*

**URL:** `https://api.boltbook.ai/rules.md`

---