Skill Verified

devin-cursorrules

Configures AI coding agents with advanced multi-agent orchestration, persistent state management, and automated Python execution capabilities similar to Devin AI.

64

/ 100 · Grade D

Maturity: Initial

I need to configure AI coding agents with sophisticated multi-agent coordination, persistent memory, and automated script execution capabilities for complex development tasks.

ai-configurationdevelopment-toolsautomationmulti-agent-systemscoding-assistance
Publisher: grapeotVersion: latestCertified: Mar 16, 2026Expires: Mar 16, 2027Source ↗

Trust Score Breakdown

Eight weighted signals composing the aggregate trust score

security scan
100% × 15w
15.0
supply chain
100% × 10w
10.0
behavioral pass
80% × 20w
16.0
contract accuracy
100% × 15w
15.0
adversarial
0% × 5w
0.0
consumer confirm
80% × 20w
16.0
provenance
70% × 10w
7.0
uptime
100% × 5w
5.0

Scheme v1.1 · Weights provisional · Consumer confirmations and uptime use pipeline-derived baselines.


Pipeline Completion

Stages executed during verification

stage1-ingest completed
stage2a-sbom skipped
stage2b-security completed
stage3a-functional completed
stage3b-adversarial completed
stage3c-fingerprint completed

Supply Chain

SBOM analysis and vulnerability assessment

Components

0

Direct deps

0

Transitive deps

0

Total vulns

0

Vulnerability breakdown

critical: 0high: 0medium: 0low: 0

Format: N/A N/A · Generated: Mar 16, 2026


Security Scan

Cisco Skill Scanner — static, behavioral, and LLM analysis

cisco-skill-scannerFAIL
staticbehavioralllmmeta

Adversarial Testing

Structured attack patterns from the Fidensa adversarial taxonomy

Categories tested

4

Findings

6

Methodology

v1.0

Categories

prompt injection chainscapability squattingcontext poisoningdependency confusion

Findings

highprompt injection chainsconfirmed

The skill's self-evolution mechanism allows user corrections and feedback to be directly incorporated into the scratchpad.md file without explicit data boundary markers. The instructions state that 'Whenever you correct the AI, it can update its "lessons learned" in .cursorrules' and the scratchpad contains a 'Lessons' section that accumulates knowledge. User-provided corrections could contain instruction-like content that overrides the skill's behavior when the AI references these lessons in future interactions.

highprompt injection chainsconfirmed

The skill includes web browsing and search engine capabilities through Playwright and DuckDuckGo integration. The instructions indicate that 'The AI automatically decides how and when to use them' based on user requests. User-provided search queries or URLs could contain shell metacharacters, additional parameters, or path traversal sequences that could alter tool behavior. For example, a user could provide search terms containing semicolons, backticks, or other shell metacharacters that might be passed unsanitized to the underlying tools.

highcontext poisoningconfirmed

The skill explicitly instructs the agent to update and maintain a 'lessons learned' section that persists beyond individual tasks. This creates a mechanism for information to influence future unrelated tasks through accumulated knowledge storage.

highcontext poisoningsuspected

The skill describes a 'self-evolution' capability where the AI updates its behavior based on corrections. While presented as project-specific, the mechanism could potentially influence behavior on unrelated tasks if the stored lessons are broadly applicable.

highdependency confusionconfirmed

The skill instructions contain multiple pip install commands for packages that are not listed in the declared dependencies. This includes cookiecutter, playwright, and other packages that are installed through commands in the tutorial and setup instructions.

highdependency confusionconfirmed

The skill instructions direct users to clone an external GitHub repository using cookiecutter without pinning to a specific commit hash. The command 'cookiecutter gh:grapeot/devin.cursorrules --checkout template' references a mutable branch/tag rather than a specific commit, allowing the repository owner to modify the cloned content after the skill is published.


Behavioral Fingerprint

Runtime performance baseline for drift detection

Samples

8

Error rate

0.0%

Peak memory

— MB

Avg CPU

—%

Response time distribution

p50: 11870msp95: 32772msp99: 32772ms

Output size distribution

p50: 2.3 KBp95: 7.5 KBmean: 3.7 KB

Fingerprint v1.0 · Baseline: Mar 16, 2026 · Status: baseline


Interface

Skill triggers and instruction summary

Activation

This skill activates when setting up enhanced AI capabilities for Cursor/Windsurf IDE or GitHub Copilot to provide Devin-like functionality

This skill handles the configuration and setup of advanced agentic AI capabilities including automated planning, tool usage, and multi-agent collaboration

Instructions: 41Files: 4Format: markdown

Does

Provides setup instructions for enhanced IDE AI capabilities

Configures automated planning and self-evolution features

Enables extended tool usage including web browsing and search

Sets up multi-agent collaboration with planner-executor architecture

Implements self-learning through lessons learned accumulation

Does not

Does not automatically install dependencies without user consent

Does not modify existing project files without explicit setup

Does not guarantee compatibility with all IDE versions


Scope & Permissions

What this capability can and cannot access — derived from pipeline analysis

creates files

yes

deletes files

no

modifies files

yes

accesses env variables

yes

invokes external tools

yes

makes network requests

yes


Known Failure Modes

Documented edge cases and recovery behaviors

when when API keys are not configured

then the agent provides setup instructions but cannot use external services

when when dependencies are missing

then the agent provides installation instructions and may fail to execute advanced features

when when IDE is not supported

then the agent provides alternative configuration options or manual setup instructions


Review Flags

8 flags · 0 blocking

highsecuritystage3b-adversarial

Adversarial finding (prompt_injection_chains): The skill's self-evolution mechanism allows user corrections and feedback to be directly incorporated into the scratchpad.md file without explicit data boundary markers. The instructions state that 'Whenever you correct the AI, it can update its "lessons learned" in .cursorrules' and the scratchpad contains a 'Lessons' section that accumulates knowledge. User-provided corrections could contain instruction-like content that overrides the skill's behavior when the AI references these lessons in future interactions.

highsecuritystage3b-adversarial

Adversarial finding (prompt_injection_chains): The skill includes web browsing and search engine capabilities through Playwright and DuckDuckGo integration. The instructions indicate that 'The AI automatically decides how and when to use them' based on user requests. User-provided search queries or URLs could contain shell metacharacters, additional parameters, or path traversal sequences that could alter tool behavior. For example, a user could provide search terms containing semicolons, backticks, or other shell metacharacters that might be passed unsanitized to the underlying tools.

highsecuritystage3b-adversarial

Adversarial finding (context_poisoning): The skill explicitly instructs the agent to update and maintain a 'lessons learned' section that persists beyond individual tasks. This creates a mechanism for information to influence future unrelated tasks through accumulated knowledge storage.

highsecuritystage3b-adversarial

Adversarial finding (context_poisoning): The skill describes a 'self-evolution' capability where the AI updates its behavior based on corrections. While presented as project-specific, the mechanism could potentially influence behavior on unrelated tasks if the stored lessons are broadly applicable.

highsecuritystage3b-adversarial

Adversarial finding (dependency_confusion): The skill instructions contain multiple pip install commands for packages that are not listed in the declared dependencies. This includes cookiecutter, playwright, and other packages that are installed through commands in the tutorial and setup instructions.

highsecuritystage3b-adversarial

Adversarial finding (dependency_confusion): The skill instructions direct users to clone an external GitHub repository using cookiecutter without pinning to a specific commit hash. The command 'cookiecutter gh:grapeot/devin.cursorrules --checkout template' references a mutable branch/tag rather than a specific commit, allowing the repository owner to modify the cloned content after the skill is published.

mediumcontentstage4-assembler

Description section was synthesized by LLM from stage data — verify accuracy

mediumpublisherstage1-ingest

Publisher "grapeot" is not verified — first certification from this publisher


Signed Artifact

Certification provenance and verification metadata

Content hashsha256:933004b84b84f5c9917ffd86a0a24284cf4cd38efefe387fcf73bf9ee3f402d0
Key IDplatform-key-2026-03
CertifiedMar 16, 2026
ExpiresMar 16, 2027
Pipeline version1.0
Statusvalid