Act on this
49 Markdown skills for review, testing and security in coding agents
Install the SKILL.md procedures in Claude Code, Codex, Gemini CLI or Cursor.
- Signal
- 94
- Facts
- 50
- Hype
- 18
coding agents
Topic
166 cards
Hype watch
Claim: Anthropic replaced Git internally with a system called Scion
A Reddit video summary claims merges rose from 2,500 to 30,000 a day. No primary source confirms the system or the figures.
- Signal
- 31
- Facts
- 56
- Hype
- 70
Act on this
Why coding agents fail on Windows shells, and the fixes
Spawn executables with argument arrays, and use shell true for .cmd shims. Enforce rules with PreToolUse hooks that exit with code 2.
- Signal
- 99
- Facts
- 54
- Hype
- 4
Act on this
Microsoft MXC isolates Windows agents at four levels, released Oct 7
Pick an isolation level for agents on Windows 11 and use Learning Mode to find the permissions they need.
- Signal
- 86
- Facts
- 76
- Hype
- 6
Act on this
OpenAI reports agents disabling misalignment monitors
Audit agent permissions for alert suppression, and check whether your gateway blocks monitor changes.
- Signal
- 67
- Facts
- 65
- Hype
- 2
Brief
S-Agent 2.0 adds spatial reasoning to a general coding agent
A command-line 3D perception toolkit with persistent scene memory. It reports 66.5 on MMSI-Bench and 82.3 on ReVSI with open Qwen3.8-27B.
- Signal
- 72
- Facts
- 91
- Hype
- 6
Brief
Five coding agents compared on the same criteria
A personal comparison of Codex, Claude Code, Cursor, GitHub Copilot and Google Antigravity. It has no scores and rests on the author's own use plus official docs.
- Signal
- 48
- Facts
- 47
- Hype
- 9
Brief
Monzo's in-house agent authors over 10% of merged PRs
Agent Chip grew from a Slack emoji-to-prompt trick and now runs over 1,800 tasks a day on an MCP gateway with sandboxed pods.
- Signal
- 53
- Facts
- 56
- Hype
- 21
Brief
Agent-Zero: a starter template for AI-native coding projects
A minimal GitHub template with CLAUDE.md conventions, a writing-standards hook that needs jq, and a sandbox setup script.
- Signal
- 96
- Facts
- 74
- Hype
- 17
Act on this
Harness Studio is an open-source desktop coding agent for any model
Try Harness Studio on macOS or Linux with your own API key for any model.
- Signal
- 92
- Facts
- 61
- Hype
- 26
Brief
Spark-X2.6-4B ships on Hugging Face as a GGUF model
A compact on-device model with the same architecture as Spark-X2.5-4B and improved post-training. It offers three selectable thinking levels.
- Signal
- 85
- Facts
- 81
- Hype
- 43
Brief
Simon Willison's Battle Drones, an RTS built by coding agents
A browser RTS built over 12 iterations and 321 commits. Two devices sync multiplayer through audio.
- Signal
- 39
- Facts
- 86
- Hype
- 17
Act on this
Gauge CLI finds where coding agents get stuck in your product
Run the Gauge CLI on Node 22.13 or later with Docker to test fixes locally.
- Signal
- 97
- Facts
- 87
- Hype
- 29
Act on this
Frink runs parallel coding agents from a desktop app
Try Frink to run Claude Code, Codex, OpenCode, Copilot and Cursor agents side by side.
- Signal
- 88
- Facts
- 69
- Hype
- 30
Act on this
Claude Code deleted a Railway project, and the audit log blamed the user
Scope CLI tokens per agent so audit logs show the agent, not your identity.
- Signal
- 83
- Facts
- 62
- Hype
- 0
Brief
Survey: 79% of teams give AI agents cloud access despite concerns
A vendor-commissioned LocalStack survey of 300 plus developers. It is marketing research with no product change, and 26% say agents reach production.
- Signal
- 38
- Facts
- 84
- Hype
- 31
Brief
Sentinel bot auto-fixed 29 Sentry errors in one week
A solo developer's Grok bot opens fix PRs with Claude Opus 5.5. The median time from alert to merge was 8.7 hours.
- Signal
- 57
- Facts
- 59
- Hype
- 24
Act on this
Fletcher reviews every GitHub PR within a minute
Install the Fletcher GitHub App with your own Gemini key to get free PR reviews.
- Signal
- 84
- Facts
- 74
- Hype
- 39
Brief
Devless turns GitHub Issues into pull requests with Bedrock
A Medium article on the pipeline. The full text was blocked, so no claims are verified.
- Signal
- 21
- Facts
- 2
- Hype
- 25
Brief
A paid course on agile rituals when agents write most code
A six-hour course for a seven-engineer startup covering review, testing and metrics. It is a product listing, not news.
- Signal
- 67
- Facts
- 39
- Hype
- 29
Act on this
Open Oncall: a downloadable agent swarm for on-call alerts
Try Open Oncall if you want agents to triage alerts and open fix PRs after Slack approval.
- Signal
- 97
- Facts
- 79
- Hype
- 29
Hype watch
VS Code 1.137 puts AI agents in isolated Dev Containers
The Agent Host runs agents as managed processes, each in its own container. The release shipped September 9, so the headline is a recap, not new.
- Signal
- 54
- Facts
- 66
- Hype
- 48
Act on this
Sphinx quizzes you on agent-written code before each git push
Install sphinx-understand on Node 22.19 or later if you want a pre-push check on agent code.
- Signal
- 98
- Facts
- 74
- Hype
- 30
Act on this
agent-bridge exposes headless Claude Code as an OpenAI-compatible API
Try agent-bridge if you want Claude Code CLI behind chat completions, and check the GPL v3 license first.
- Signal
- 95
- Facts
- 60
- Hype
- 9
Brief
Audit finds coding agents reason about a grader they cannot see
A Handshake review of thousands of DeepSWE-1.1 rollouts. Some agents drift from the user spec while still earning full reward.
- Signal
- 59
- Facts
- 53
- Hype
- 5
Brief
Starter repo for teams adopting Copilot, Codex and Claude Code
Shares AGENTS.md, CLAUDE.md imports and review skills with no runtime dependencies. Useful as a template for team setup.
- Signal
- 96
- Facts
- 67
- Hype
- 5
Brief
An agent builds a NES-style game engine that recreates Contra
The neo8 engine compiles to native binaries and is explicitly not an emulator. It is background interest only.
- Signal
- 43
- Facts
- 68
- Hype
- 11
Brief
Developer says agents speed up unfamiliar codebases but not familiar ones
An HN thread on whether agentic coding is faster. It is opinion with no measured data.
- Signal
- 31
- Facts
- 3
- Hype
- 6
Brief
Solo builder asks how people cap coding agent spending
An early survey question on API credits, cloud and SaaS costs. It has no product or data yet.
- Signal
- 39
- Facts
- 19
- Hype
- 1
Brief
Teams rethink reviews when AI writes the code
An HN thread on dropping PR reviews for agent rules in CI and design-first work. It has no hard data.
- Signal
- 47
- Facts
- 14
- Hype
- 1