Back to skill

Security audit

Spm V4 Push

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a coherent project-management tool, but it under-discloses broad Git commit behavior and overstates a weak command security gate.

Install only if you are comfortable with a project-management skill that creates local tracking files and may direct subagents to make commits. Review or remove the implementer prompt's git add -A/commit step, and do not rely on the security gate as a complete command sandbox.

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 (2)

T09 · Insecure Skill Coding Practices

Error
Location
src/security/classifier.js:55
Finding

Security Gate Fails Open and Is Trivially Bypassable

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
prompts/implementer.md:58
Finding

Implementer Prompt Indiscriminately Stages Unrelated Repository Content

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • 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
Findings (127)

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The declared description describes a full-featured project management engine with multiple substantive capabilities. The supplied code chunk, however, is only a lightweight export barrel for CLI command handlers. While some exported command names like 'attest' and 'verify' may suggest related functionality exists elsewhere, this specific code does not implement those features and its actual purpose is limited to organizing exports. That is a materially different primary purpose from the declared description, so this chunk does not accurately represent the stated functionality.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared description claims a full-featured project management engine with multiple concrete subsystems and orchestration capabilities. The supplied code chunk only provides a handler type definition/stub for CLI commands and explicitly states that the function body is not used. It does not implement or expose any of the declared engine features, nor any substantive runtime behavior beyond throwing an error if invoked. This is a material description-behavior mismatch, not merely an incomplete supporting detail.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The declared description presents a broad, production-grade project management engine with multiple substantial capabilities. The supplied code chunk implements only a specific CLI quality-check command over a WBS ledger. It validates task metadata, reports pass/warn/fail results, and optionally records an event. While this aligns partially with the 'WBS ledger' and possibly event-store aspects of the description, it does not implement or demonstrate most of the headline capabilities: no 6-phase state machine, no SHA-256 attestation, no YAML security gate, and no sub-agent orchestration. The primary purpose of this code is narrower than the declared description, so the description does not accurately represent what this chunk actually does.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The supplied code chunk implements only a narrow integrity-verification CLI for a WBS ledger and its SHA-256-style attestation record. This partially aligns with the declared mention of a WBS ledger and attestation, but it does not demonstrate the broader claimed behavior of a full project management engine, 6-phase state machine, YAML security gate, or sub-agent orchestration. It also only optionally appends a single integrity event if a local event-store directory exists, which is much narrower than the declared 3-domain event store capability. Because the description presents a substantially broader primary purpose and capabilities than this code actually performs, the description does not accurately represent this code chunk.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The code chunk is narrowly scoped to configuration loading. It defines default configuration values that reference lifecycle phases, event-store domains, security policy path, and WBS settings, but it does not actually implement those systems or behaviors. There is no project management engine logic, no orchestration, no hashing/attestation, no event persistence, and no security gate enforcement. While the defaults mirror the declared product description, the actual behavior of this code is only config parsing, validation, and merging, which is a materially narrower and different purpose than the declared description.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The supplied code chunk does accurately implement part of the description: a 6-phase workflow/state machine with events, transition checks, context, and history. However, the declared purpose is much broader and includes several concrete capabilities not present in this code: no WBS ledger, no cryptographic attestation, no event store architecture beyond in-memory listeners/history, no YAML security gate, and no sub-agent orchestration. The primary behavior of this chunk is a state machine engine, not the full production-grade project management system described. Therefore the description materially overstates what this code actually does.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The supplied code chunk partially matches the description only in that it implements the 6-phase state machine portion of the system. It defines phase metadata, transition rules, preconditions/postconditions, and helper functions for building and validating phase definitions. However, none of the other prominently declared capabilities are present in this code: there is no work-breakdown structure ledger, no cryptographic SHA-256 attestation, no event store, no YAML-based security gate, and no sub-agent orchestration. Because the declared purpose presents a broader production-grade engine while the actual code is limited to phase-definition logic, the description does not accurately represent what this chunk actually does.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied code only provides a workflow/state progression helper and registry around an existing Engine interface. It supports phase sequencing, trigger registration, simple condition evaluation, and serialization of workflow metadata/state. While this partially aligns with a generic 'phase state machine' concept, it does not implement or demonstrate the major declared capabilities: no WBS ledger, no cryptographic SHA-256 attestation, no YAML parsing or security gate, no event-store implementation (let alone a 3-domain one), and no sub-agent orchestration. The description therefore materially overstates what this code chunk actually does.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The code chunk is a narrow utility for constructing event-store domain configuration metadata. It does align with one small part of the description—the existence of a 3-domain event store (audit, integrity, quality)—but it does not implement or demonstrate the broader declared purpose of a production-grade AI coding project management engine. There is no evidence here of project/task management, workflow state transitions, attestation computation, security gate enforcement, or orchestration. Because the declared description presents a much broader and materially different capability set than this code actually provides, this is a mismatch.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied code chunk is narrowly focused on event-store functionality: creating a data directory, validating events, appending them to per-domain JSONL files, querying/filtering, and maintenance operations like repair/rotate/prune. While it references YAML parsing via import, there is no visible YAML-based security gate logic. The broader declared description describes a much larger project-management/orchestration system with attestation, workflow state machine, and sub-agent features, none of which are implemented in this code. Therefore the description materially overstates what this code chunk actually does.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied code chunk is narrowly focused on event log file storage and maintenance. It appends JSONL events with locking, computes and verifies a truncated SHA-256 checksum, reads and repairs logs, rotates archives, and prunes old files. While this partially aligns with the 'event store' and 'SHA-256 attestation' aspects of the description, it does not implement the broader claimed functionality of a 'Super Project Manager' engine, including WBS management, phase/state-machine control, YAML security gating, or sub-agent orchestration. Additionally, the code performs concrete filesystem maintenance actions such as rotation and pruning (deleting old log files), which are not mentioned in the declared description. Therefore the description materially overstates and misrepresents what this code chunk actually does.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The supplied code chunk is narrowly focused on hook management infrastructure. It supports plugin-style interception around tool calls and phase transitions, which could be a supporting component of a larger project manager, but it does not itself implement most of the prominently declared capabilities. The only partial alignment is with phase-related events and extensibility hooks that might fit into a state machine/orchestration system. However, there is no evidence here of a WBS ledger, attestation/hashing, event store, YAML-based security checks, or sub-agent orchestration. Therefore the description materially overstates what this code chunk actually does.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The supplied code chunk only provides WBS context injection for downstream tool handlers. It reads ledger state, formats active/completed/pending tasks, and attaches that string to context on preToolUse. While this is consistent with a small part of the declared WBS-related functionality, it does not implement or evidence most of the prominently declared capabilities such as attestation, event store, security gate, state machine, or orchestration. The description therefore materially overstates what this code chunk actually does.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The supplied code chunk does not implement the broad project-management/orchestration engine described. It only performs regex-based classification of shell command strings against provided policy rules, returning a verdict and defaulting to safe when no rule matches. While this could plausibly be one small part of the declared 'YAML security gate,' the declared description emphasizes a much larger and materially different primary purpose with multiple major capabilities that are absent from this code. Therefore the description does not accurately represent what this specific code chunk actually does.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The supplied code chunk is narrowly focused on a security gate for shell-command classification using a policy engine. While this partially aligns with the declared mention of a 'YAML security gate,' it does not implement or evidence the larger claimed system: project management engine behavior, ledger/state-machine logic, attestation, event store, or orchestration. Therefore the description materially overstates and misrepresents the primary purpose of this code chunk.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

This code chunk partially aligns with the declared mention of a 'YAML security gate' because it loads YAML policy rules and provides built-in command-classification rules. However, the declared description presents a broad AI coding project management engine with multiple major subsystems, while the supplied code only implements one small security-policy component. There is no evidence here of project management logic, ledgering, attestation, event storage, orchestration, or a phase state machine. Thus the description does not accurately represent what this specific code chunk actually does.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The supplied code chunk is a small session heartbeat persistence module. It resolves a local file path, ensures the parent directory exists, appends heartbeat JSON lines, reads all entries, returns the latest entry, and filters by session ID. This is consistent with session tracking, but it does not implement or demonstrate the major capabilities claimed in the declared description, such as project management orchestration, attestation, security gating, or multi-domain event storage. The mismatch is therefore material: the declared purpose describes a broad production-grade engine, while the actual code only provides a supporting heartbeat logging component.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The declared description presents a broad, production-grade project management and orchestration engine with multiple security and workflow capabilities. The supplied code chunk does not implement those capabilities. Instead, it only reads heartbeat data, determines checkpoints, summarizes session state, and looks up heartbeats by time. This is a materially narrower and different function than the declared purpose, so the description does not accurately represent the code.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The supplied code chunk does not implement the broad engine described. It contains only data validation schemas: project name/path input, ledger path config, policy rules, task objects, and event objects. While there is partial alignment with concepts like a WBS/WBT task model, ledger path configuration, policy rules, and a 3-domain event structure, the chunk does not implement a 6-phase state machine, SHA-256 attestation, YAML parsing/gating, or any sub-agent orchestration. Therefore the declared description materially overstates what this code actually does.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The supplied code matches one narrow part of the declared description: SHA-256 attestation for WBS ledger integrity. It computes hashes, stores an attestation record on disk, loads it, and verifies integrity. However, the declared purpose describes a much broader system with multiple major capabilities (state machine, event store, YAML security gate, orchestration) that are not present in this code chunk. Since the evaluation is description versus this supplied code chunk, the description materially overstates the actual behavior of the code shown.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The supplied code chunk is narrowly focused on WBS ledger parsing and task CRUD/validation. It does support a task state machine, but only simple task status transitions among todo/doing/done/blocked/skipped—not a clearly defined 6-phase project state machine. There is no evidence of SHA-256 hashing/attestation, event store implementation, YAML-based security gating, or any sub-agent orchestration. Therefore the description materially overstates the implemented functionality in this code chunk.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The supplied code chunk is narrowly focused on WBS task integrity hashing and serialization. It does align with part of the declared description mentioning WBS ledger / SHA-256 attestation, but it does not implement or demonstrate the broader advertised capabilities such as a 6-phase state machine, event store, YAML security gate, or sub-agent orchestration. Since the declared purpose presents the skill as a full project management engine while the actual code is only a Merkle hashing module, the description materially overstates the behavior of this code chunk.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied code does not implement the broad production-grade project management engine described. Instead, it is exclusively test code validating parts of an Engine and Workflow system. There is partial alignment on the existence of a 6-phase state machine and event hooks/workflow behavior, but the chunk does not show or exercise several headline capabilities in the description: no SHA-256 hashing/attestation, no WBS ledger, no YAML security gate, no 3-domain event store, and no sub-agent orchestration. Because the declared purpose emphasizes a much larger feature set than what this code chunk actually does, the description does not accurately represent the code shown.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The code shown does not implement or exercise a project management engine as described. It is specifically test code for an event store subsystem: parsing JSON lines, appending and querying events, rotating files, pruning old files, and handling three configured domains. The only declared feature that aligns is the existence of a 3-domain event store (audit, integrity, quality). There is no evidence in this chunk of a WBS ledger, phase/state-machine workflow, SHA-256 attestation logic, YAML-based security gating, or any sub-agent orchestration. Because the declared description presents a much broader and materially different primary purpose than the observed code behavior, this is a mismatch.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The code chunk does not implement or demonstrate the broad project-management engine described. Instead, it is narrowly focused on unit tests for hook registry behavior and a WBS context injection hook. While there is some partial alignment with hook/event behavior and WBS context injection, the major declared capabilities—attestation, event store, YAML security gate, and sub-agent orchestration—are absent. The primary purpose of this code chunk is testing middleware hooks, not operating a full project management engine as described.

Content

No source excerpt is available for this finding.