Agent Reflection Engine

Enables AI agents to self-audit decision steps, identify reasoning bottlenecks, and generate improvement patches via chain-of-thought critique.

Security Scan

The skill's code and instructions broadly match its stated purpose (analyzing agent traces), but there are multiple coherence and correctness issues in the instructions and code that could cause failures or unexpected behavior; review and fix before use, especially on sensitive traces.

Purpose & Capability

Name/description match the included code: both describe a reflection engine that reads a JSON trace and emits critiques. No unrelated credentials, binaries, or network access are requested. The inclusion of a small Python tool is proportional to the stated purpose.

Instruction Scope

SKILL.md example calls 'python agent_reflection_engine.py' but the provided file is tool.py (and the module's top comment names agent_reflection_engine.py). The code operates only on a provided trace file (no exfiltration), but there are clear logic bugs (e.g., building 'report' then referencing 'report' inside the same dict for summary — this will raise a NameError at runtime). The reflection heuristics are also naive (misspelled substring checks like 'inconsisten' and 'efficien'), which could produce unhelpful or misleading output. These inconsistencies mean the runtime behavior may differ from the documented usage and could crash.

Install Mechanism

No install spec — instruction-only plus a single Python file. No external downloads or package installs are requested, which minimizes supply-chain risk.

Credentials

No environment variables, credentials, or config paths are required. The tool operates on a local JSON trace file only, which is proportionate to the stated purpose.

Persistence & Privilege

The skill does not request persistent/always-on presence or privilege escalation. It is user-invocable and does not alter other skills or system-wide settings.

What to consider before installing

This skill is not obviously malicious, but it contains coherence and correctness problems you should fix before trusting it with real traces. Recommended steps: (1) Run it in a sandbox with non-sensitive demo traces to reproduce behavior. (2) Rename or update invocation examples so SKILL.md matches the actual filename (tool.py) or rename the file. (3) Fix the runtime bug where the report's summary references 'report' while the dict is being built (this currently causes a crash). (4) Review and improve the simplistic string checks (misspellings like 'inconsisten'/'efficien' make the heuristics unreliable). (5) Only run this on sensitive agent traces after the above fixes and after verifying output does not leak data externally. If you need higher assurance, request the author to provide a corrected release and unit tests or perform a code review focused on correctness and privacy.

A lightweight, pluggable reflection engine that enables AI agents to self-audit their decision traces, identify reasoning bottlenecks, and generate improvement patches using chain-of-thought critique—ideal for developers tuning autonomous agents.

Usage

# Run reflection on an agent trace
python agent_reflection_engine.py traces/demo_trace.json -o reports/reflection.json --verbose

# Example trace format (demo_trace.json):
# [
#   {
#     "step_id": 1,
#     "thoughts": "I should search for the nearest coffee shop.",
#     "action": "search_web",
#     "value": "coffee shop near me",
#     "observation": "Found 'Brew Haven' 0.3 miles away."
#   }
# ]

Integrate into agent loops by logging each step and running periodic reflection to generate improvement heuristics.

Price

$4.99