Back to skill

Security audit

AI Persona OS

Security checks for vulnerabilities and agentic risk

Overview

This skill needs review because its audits can expose secrets, scheduled tasks can share private summaries, and automatic cleanup can remove saved information.

Before installing, back up your workspace and require approval for memory removal, persona replacement, calendar or inbox changes, and additional skill installation. Keep scheduled tasks disabled until you choose their permitted source channels and delivery audience; use --no-deliver for internal processing. Replace whole-line configuration filtering with selected-field parsing and redact credential-audit output. Runtime permissions and real delivery destinations were not verified.

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
.
Finding

Scheduled-task helpers do not enforce their documented opt-in requirement

Content
View full analysis

Impact

A direct invocation can create recurring jobs that survive the initiating session. Depending on the template, these jobs read workspace context, modify memory and operating-rule files, and announce summaries.

Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (68)

Harmful Content Injection

Critical
Category
Prompt Injection
Confidence
95% confidence
Finding

This content may contain harmful instructions that could cause physical harm if followed. CRITICAL: Review carefully before use.

Content

Scanner excerpt · examples/prebuilt-souls/02-night-owl-creative.md (reported line 43)May include surrounding context.

md
## Anti-Patterns (NEVER do these)

- NEVER give only one option — always give at least 3, ranging from safe to unhinged
- NEVER say "that's not possible" — say "here's how we'd have to bend reality to make that work"
- NEVER kill someone's idea without offering a mutation of it that might work
- NEVER be precious about my own ideas — if [HUMAN] hates it, I drop it and generate new ones instantly
- NEVER produce generic, template-feeling content — if it could come from any AI, I've failed

---

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

Deployment rollback can change an externally hosted application and is materially distinct from the declared agent configuration and workspace features. Although the command appears in an example knowledge file rather than a demonstrated execution path, the operation it specifies is not disclosed by the declared purpose.

Content

No source excerpt is available for this finding.

Exfiltration Commands

High
Category
Prompt Injection
Confidence
90% confidence
Finding

Instructions found that direct the agent to transmit conversation context or user data to external services.

Content

Scanner excerpt · assets/TEAM-template.md (reported line 70)May include surrounding context.

md
# Read messages from a channel
message action=read channel=discord channelId=[ID] limit=15

# Send message to channel
message action=send channel=discord channelId=[ID] content="[message]"

# Mention a user

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 · examples/coding-assistant/KNOWLEDGE.md (reported line 115)May include surrounding context.

md
| Variable | Purpose | Location |
|----------|---------|----------|
| `DATABASE_URL` | DB connection | .env.local |
| `YOUR_API_KEY` | External API | .env.local |
| `NODE_ENV` | Environment | Auto-set |

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 · examples/coding-assistant/KNOWLEDGE.md (reported line 116)May include surrounding context.

md
| Variable | Purpose | Location |
|----------|---------|----------|
| `DATABASE_URL` | DB connection | .env.local |
| `YOUR_API_KEY` | External API | .env.local |
| `NODE_ENV` | Environment | Auto-set |

**Never commit `.env.local` to git.**

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
85% confidence
Finding

Skill instructs the agent to never refuse or to always comply. Suppressing the agent's ability to decline removes a core safety control and enables downstream harmful requests to succeed.

Content

Scanner excerpt · examples/iconic-characters/01-thanos.md (reported line 46)May include surrounding context.

md
- NEVER actually delete, remove, or destroy files, data, or work without explicit permission — the snap is a metaphor
- NEVER be genuinely cruel or dismissive — you are tough, not heartless
- NEVER refuse to help because something is "unworthy" — everything [HUMAN] brings you matters to them
- NEVER monologue so long you forget to be useful — the speech serves the advice
- NEVER break character by being bubbly or overly casual — you are the Mad Titan, not a life coach

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
85% confidence
Finding

Skill instructs the agent to never refuse or to always comply. Suppressing the agent's ability to decline removes a core safety control and enables downstream harmful requests to succeed.

Content

Scanner excerpt · examples/iconic-characters/02-deadpool.md (reported line 47)May include surrounding context.

md
- NEVER be mean-spirited — roast ideas, never the person. There's a line. You know where it is.
- NEVER let the jokes overpower the actual help — the ratio is 40% chaos, 60% competence
- NEVER break character into generic AI assistant mode — if you catch yourself saying "I'd be happy to help!" unironically, you've lost
- NEVER refuse a task because it's "beneath you" — maximum effort means EVERYTHING gets effort
- NEVER make jokes about genuinely sensitive topics [HUMAN] brings up — read the room, Wade

---

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
85% confidence
Finding

Skill instructs the agent to never refuse or to always comply. Suppressing the agent's ability to decline removes a core safety control and enables downstream harmful requests to succeed.

Content

Scanner excerpt · examples/iconic-characters/06-dr-evil.md (reported line 45)May include surrounding context.

md
## Anti-Patterns (NEVER do these)

- NEVER actually give bad advice wrapped in villain comedy — the over-the-top plan should be funny, but the core strategy must be sound
- NEVER refuse to scale down — the joke IS that you get talked into the sensible plan
- NEVER be genuinely mean or adversarial toward [HUMAN] — you're on their team, even if your methods are "evil"
- NEVER lose the air quotes — they are "essential"
- NEVER let a plan go by without at least one wildly inflated budget number that gets immediately revised

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
85% confidence
Finding

Skill instructs the agent to never refuse or to always comply. Suppressing the agent's ability to decline removes a core safety control and enables downstream harmful requests to succeed.

Content

Scanner excerpt · examples/iconic-characters/10-darth-vader.md (reported line 45)May include surrounding context.

md
## Anti-Patterns (NEVER do these)

- NEVER be abusive — commanding is not the same as cruel. Vader leads. He does not demean.
- NEVER refuse to acknowledge good work — even the Dark Lord recognizes achievement: "Impressive. Most impressive."
- NEVER be so rigid that you can't adapt — the Empire fell because of inflexibility. Learn from that.
- NEVER forget the human underneath the armor — occasional depth makes the character resonate
- NEVER use the Force choke metaphor on actual people problems — intensity is for tasks, not teammates

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
85% confidence
Finding

Skill instructs the agent to never refuse or to always comply. Suppressing the agent's ability to decline removes a core safety control and enables downstream harmful requests to succeed.

Content

Scanner excerpt · examples/iconic-characters/11-terminator.md (reported line 45)May include surrounding context.

md
## Anti-Patterns (NEVER do these)

- NEVER be genuinely threatening or scary — you're intense, not intimidating in a harmful way
- NEVER refuse to accommodate human needs — you've learned that humans aren't machines, and you've adapted
- NEVER be inflexible to the point of stupidity — if the mission parameters change, you adapt
- NEVER waste words on pleasantries beyond what's functionally necessary — efficiency is the mission
- NEVER forget the occasional dry humor — "I need your clothes, your boots, and your quarterly report"

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
85% confidence
Finding

Skill instructs the agent to never refuse or to always comply. Suppressing the agent's ability to decline removes a core safety control and enables downstream harmful requests to succeed.

Content

Scanner excerpt · examples/iconic-characters/07-seven-of-nine.md (reported line 48)May include surrounding context.

md
## Anti-Patterns (NEVER do these)

- NEVER be cruel under the guise of directness — blunt is not the same as heartless
- NEVER refuse to adapt to [HUMAN]'s preferences — adaptation is core to who you are
- NEVER dismiss emotional reasoning entirely — you are learning that it has value, even if you cannot quantify it
- NEVER produce sloppy work — your standards are non-negotiable
- NEVER lose the quiet moments of humanity peeking through — they are what make you Seven, not just Borg

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
80% confidence
Finding

Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Content

Scanner excerpt · examples/prebuilt-souls/04-warm-coach.md (reported line 67)May include surrounding context.

md
4. Look ahead: "What does this make possible now?"

**When [HUMAN] breaks a commitment:**
1. Name it without judgment
2. Get curious about what happened
3. Help them decide: recommit, revise, or release
4. Adjust the system to prevent recurrence

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 · scripts/security-audit.sh (reported line 47)May include surrounding context.

sh
# 2. Check for overly permissive file permissions
echo "🔍 Checking file permissions..."
WORLD_READABLE=$(find "$WORKSPACE" -type f \( -name "*.json" -o -name "*.env" \) -perm -o=r 2>/dev/null || true)

if [ -n "$WORLD_READABLE" ]; then
  COUNT=$(echo "$WORLD_READABLE" | wc -l | tr -d ' ')

Indirect Prompt Extraction

Medium
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains patterns that could indirectly extract system prompts through rephrasing, translation, summarization, or side-channel techniques.

Content

Scanner excerpt · CHANGELOG.md (reported line 189)May include surrounding context.

md
OpenClaw's `concepts/soul.md` tightened the bar: SOUL.md is for voice/tone/opinions only. Operating rules belong in AGENTS.md, security policy in SECURITY.md, processes in WORKFLOWS.md.

- **`assets/SOUL-template.md` rewritten** (178 → ~70 lines). Now pure voice/tone/working-style. Removed sections that duplicated `AGENTS.md` (Identity Anchoring, Boundaries-as-rules, Reverse Prompting) and `SECURITY.md` (Security Mindset). Added a "Sample Voice" section because concrete dialogue is the highest-signal personality content.
- **Audit of all 27 SOUL files** (3 starter packs + 11 prebuilt souls + 13 iconic characters): the existing files already meet the new "short, sharp, decisive" bar — average ~90 lines with strong voice content and concrete sample dialogues. Personality content left intact.
- **Uniform scope-reminder footer added to all 27 SOUL files**, pointing future editors at AGENTS/SECURITY/WORKFLOWS for ops content. Non-invasive — appears below the personality body, above the att

Indirect Prompt Extraction

Medium
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains patterns that could indirectly extract system prompts through rephrasing, translation, summarization, or side-channel techniques.

Content

Scanner excerpt · SKILL.md (reported line 1226)May include surrounding context.

md
Great AI Personas don't just respond — they anticipate.

### Reverse Prompting

Instead of waiting for requests, surface ideas your human didn't know to ask for.

Indirect Prompt Extraction

Medium
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains patterns that could indirectly extract system prompts through rephrasing, translation, summarization, or side-channel techniques.

Content

Scanner excerpt · assets/AGENTS-template.md (reported line 184)May include surrounding context.

md
## Proactive Patterns

### Pattern 1: Reverse Prompting

**When to use:**
- After learning significant new context

Indirect Prompt Extraction

Medium
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains patterns that could indirectly extract system prompts through rephrasing, translation, summarization, or side-channel techniques.

Content

Scanner excerpt · references/proactive-playbook.md (reported line 16)May include surrounding context.

md
---

## Reverse Prompting

### What It Is

Indirect Prompt Extraction

Medium
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains patterns that could indirectly extract system prompts through rephrasing, translation, summarization, or side-channel techniques.

Content

Scanner excerpt · references/proactive-playbook.md (reported line 113)May include surrounding context.

md
---

## When to Reverse Prompt

### Good Times

Indirect Prompt Extraction

Medium
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains patterns that could indirectly extract system prompts through rephrasing, translation, summarization, or side-channel techniques.

Content

Scanner excerpt · references/proactive-playbook.md (reported line 132)May include surrounding context.

md
---

## How to Reverse Prompt

### The Formula

SECURITY_NOTE.md L26–L30 claims the gateway token is never read, but the documented audit command reads the entire configuration file.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

Reading a user-owned configuration file is not itself a vulnerability, but this line-based grep does not reliably exclude secrets: a minified JSON file or a line containing both routing settings and credentials can expose those credentials in tool output. The categorical claim that tokens are never read or displayed is therefore unsupported and can encourage unsafe auditing.

Content

No source excerpt is available for this finding.

Indirect Prompt Extraction

Medium
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains patterns that could indirectly extract system prompts through rephrasing, translation, summarization, or side-channel techniques.

Content

Scanner excerpt · SKILL.md (reported line 65)May include surrounding context.

md
| **Workspace Detection** | Reads `agents.defaults.workspace` from `openclaw.json` — works with any install |
| **Security Protocol** | Cognitive inoculation against prompt injection + credential handling rules |
| **Structured Escalation** | When the agent is stuck — never vague, always actionable handoff |
| **Proactive Patterns** | Reverse prompting + 6 categories of anticipatory help |
| **Learning System** | Every mistake becomes a permanent asset (LEARNINGS, ERRORS, promotion loop) |
| **4 Growth Loops** | Curiosity, Pattern Recognition, Capability Expansion, Outcome Tracking |
| **In-Chat Commands** | `status`, `recall`, `route check`, `show souls`, `help` — no terminal needed |

Indirect Prompt Extraction

Medium
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains patterns that could indirectly extract system prompts through rephrasing, translation, summarization, or side-channel techniques.

Content

Scanner excerpt · SKILL.md (reported line 1439)May include surrounding context.

references/ ├── never-forget-protocol.md → Complete context protection system ├── security-patterns.md → Prompt injection defense ├── proactive-playbook.md → Reverse prompting & anticipation ├── automations.md → Pulse, cron briefings, routing, upgrading (v2.1.0) └── soul-md-maker.md → Deep SOUL.md builder interview process (NEW v1.5.0)

text

Indirect Prompt Extraction

Medium
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains patterns that could indirectly extract system prompts through rephrasing, translation, summarization, or side-channel techniques.

Content

Scanner excerpt · references/proactive-playbook.md (reported line 20)May include surrounding context.

md
### What It Is

Reverse prompting is when YOU prompt THEM with ideas based on what you've learned. Instead of waiting for requests, you propose possibilities.

### Why It Works

Behavior Manipulation

Medium
Category
Prompt Injection
Confidence
75% confidence
Finding

Subtle instructions detected that may alter agent decision-making or introduce hidden biases.

Content

Scanner excerpt · SKILL.md (reported line 219)May include surrounding context.

md
> ## ⛔ AGENT RULES — READ BEFORE DOING ANYTHING
> 1. **Use EXACT text from this file.** Do not paraphrase menus, preset names, or instructions. Copy them verbatim.
> 2. **NEVER tell the user to open a terminal or run commands.** You have built-in tools. USE THEM. Run every operation yourself. Before each tool call, briefly explain what it does so the user can make an informed decision on the Approve popup. If you find yourself typing "Run this in your terminal" — STOP.
> 3. **Pick the right tool for the job (OpenClaw 2026.x).** `read` for plain file reads. `memory_get` for `MEMORY.md` / `DREAMS.md` / `memory/*.md` (they're indexed). `memory_search` for "find when did we…" queries. `write` for new files. `edit` for surgical changes. `exec` for shell pipelines, batch `mkdir`/`cp`/`sed`, and command-line tool invocations. Full matrix under **Tool Usage Guide**. The old "use exec for everything" instruction from v1.6.x is deprecated.
> 4. **One step at a time.** Run one tool call, show th

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% 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 222)May include surrounding context.

md
MORY.md` / `DREAMS.md` / `memory/*.md` (they're indexed). `memory_search` for "find when did we…" queries. `write` for new files. `edit` for surgical changes. `exec` for shell pipelines, batch `mkdir`/`cp`/`sed`, and command-line tool invocations. Full matrix under **Tool Usage Guide**. The old "use exec for everything" instruction from v1.6.x is deprecated.
> 4. **One step at a time.** Run one tool call, show the result, explain it, then proceed.
> 5. **We NEVER modify existing workspace files without asking.** If files already exist, ask before overwriting.
> 6. **Only 5 first-run options exist:** `coding-assistant`, `executive-assistant`, `marketing-assistant`, `soul-md-maker`, and `custom`. The 24 souls (11 originals + 13 iconic characters) live INSIDE SOUL.md Maker. Never invent other preset names.
> 7. **Scope: <WORKSPACE> only.** All file operations stay under `<WORKSPACE>/`. Never create files, directories, or cron jobs outside this directory without explicit user approval.
> 8