Releases·Tool Release
Dev ships cogext-scan, a zero-dependency CLI that AST-scans LangChain and CrewAI code for silent failures and runaway loops
A developer open-sourced cogext-scan, a zero-dependency Python CLI that uses static AST parsing to flag failure modes in agent code before deployment. It was motivated by production incidents where agents reported success despite failing tools, such as a SendGrid wrapper returning "Email sent" on a 401 Unauthorized. The tool runs offline, requires no account, and outputs an Agent Safety Index score.
1 engagement1 sourceWed, Sep 30, 2026, 12:11 PM
AI Overview
- cogext-scan, built after a custom tool masked SendGrid 401 errors with 'Email sent', statically parses agent code offline via Python AST
- The CLI flags try/except blocks returning static success strings, AgentExecutor or Crew instances lacking max_iterations or cost caps, and unvalidated delete_* or drop_* functions
- It also detects hardcoded API key strings, runs locally with no account via pip install cogext-scan, and outputs an Agent Safety Index score
- The open-source code is at github.com/yaminbinyoosuf/cogext-scan; the developer seeks feedback on additional AST failure patterns
Related
Releases·Tool Release
Agent config linter catches vague tool descriptions before runtime
A developer built a deterministic linter for agent configs to catch vague tool descriptions, overlapping tool boundaries, missing stopping conditions, and schema disagreements before runtime. The linter makes no model calls and was adopted by Character.AI's Larch framework, scanning 236 agent and skill files in CI.
A developer built a deterministic linter for agent configs to catch vague tool descriptions, overlapping tool boundaries, missing stopping conditions, and schema disagreements before runtime. The linter makes no model calls and was adopted by Character.AI's Larch framework, scanning 236 agent and skill files in CI.
7 engagement1 sourceredditTue, Aug 4, 2026, 05:13 PM
Releases·Tool Release
Show HN: Sensez – helping coding agents catch their own code smells
A developer released Sensez on GitHub, a tool designed to help coding agents detect code smells like dict[str, Any], boolean flag proliferation, and untestable nested functions that AGENTS.md instructions fail to prevent.
A developer released Sensez on GitHub, a tool designed to help coding agents detect code smells like dict[str, Any], boolean flag proliferation, and untestable nested functions that AGENTS.md instructions fail to prevent.
1 engagement1 sourcehackernewsWed, Sep 2, 2026, 02:29 PM
Products·building
Scanner that finds agent failures repeating across runs by reading session logs
A tool that scans agent session logs to detect external tool failures (like repeated 429 errors) that show up across multiple runs. Because agents start from nothing each run, they rediscover fixes repeatedly; small local models may silently retry and move on. Aimed at developers running scheduled or repeated agent jobs like nightly scrapes and queue workers.
A tool that scans agent session logs to detect external tool failures (like repeated 429 errors) that show up across multiple runs. Because agents start from nothing each run, they rediscover fixes repeatedly; small local models may silently retry and move on. Aimed at developers running scheduled or repeated agent jobs like nightly scrapes and queue workers.
17 engagement1 sourceredditTue, Sep 15, 2026, 06:47 PM
Releases·Tool Release
Gorgona: An open source guardrail engine that turns runaway AI agents to stone (zero dependencies)
A solo founder open-sourced Gorgona, a Python guardrail engine that halts misbehaving autonomous agents in real time. It detects leaked API keys, cross-project context leaks, and recursive loops, and halts execution immediately on hard violations. Built from the author's own experience running coding and marketing agents without safeguards.
A solo founder open-sourced Gorgona, a Python guardrail engine that halts misbehaving autonomous agents in real time. It detects leaked API keys, cross-project context leaks, and recursive loops, and halts execution immediately on hard violations. Built from the author's own experience running coding and marketing agents without safeguards.
4 engagement1 sourceredditThu, Sep 3, 2026, 06:58 AM
Products·building
SafeAI: Static analyzer for AI applications
SafeAI is an open-source static analyzer that scans AI application codebases to extract AI-specific artifacts like agent capabilities, tool definitions, prompts, and model configurations before runtime. It helps developers understand and inventory AI components in their projects, aiding in security and compliance.
SafeAI is an open-source static analyzer that scans AI application codebases to extract AI-specific artifacts like agent capabilities, tool definitions, prompts, and model configurations before runtime. It helps developers understand and inventory AI components in their projects, aiding in security and compliance.
5 engagement1 sourceredditTue, Aug 4, 2026, 10:30 PM