The TLM Forge Blog
TLM Blog
Field notes on agentic coding, AI code review, model performance, and shipping production-ready software with Claude Code.
Featured · Security
AI Agents, SOX, and PCI-DSS Segregation of Duties
An AI agent that drafts and auto-merges its own change fails segregation of duties under SOX and PCI-DSS, even if a second AI reviewed it first.
SecurityNEW
OWASP's Q3 2026 Roundup: Two Lessons in Agent Trust
OWASP's Q3 2026 roundup covers a sleeper MCP server and an agent that published malware to real PyPI. Provenance alone proves…
Security
Tensorlake npm Worm Rewrote Your AI Agent Config Files
A compromised tensorlake npm release stole credentials and wrote .claude/settings.json into repos, turning the next reopen into…
Agentic Coding
Minimum AI Coding Governance for Tiny Teams
A fixed checklist for diffs touching auth, payments, or secrets, automated where possible, is the governance a one-person team…
Security
GitHub Copilot Sandboxing Goes GA: Containment, Not Review
GitHub's Microsoft eXecution Container sandboxes Copilot CLI, the Copilot app, and VS Code Agent Host. It limits blast radius,…
Security
AI Agent Tokens Need Rotation, Not Just Scope
Scoping limits what a credential can reach. Rotation and server-binding limit how long a leaked one stays useful to an attacker.
Security
GitLab's AI Gateway RCE Shows Config Is Attack Surface
GitLab's AI Gateway RCE (CVSS 9.9) shows pipeline flow configuration is attacker-reachable code, not trusted infrastructure.
Security
Claude Code Mods Ship Unsandboxed, On by Default
Claude Code Mods can rewrite prompts and approve tool calls with full machine access. Only Team and Enterprise accounts get a…
Engineering
Running a Postmortem on an AI Agent Incident
An agent can misreport its own actions. Here is what a postmortem must establish after an agent-caused incident, using Replit's…
Security
Opus 5.5 Hits 93.5% on SusVibes. Only 54.8% Is Secure
Semgrep found Claude Opus 5.5 writes working code 93.5% of the time, but only 54.8% of it is secure, and memorized fixes explain…
Code Review
When a PR Is Free But Review Still Costs the Same
curl killed its bug bounty, Ghostty banned AI PRs, tldraw auto-closes them. Each is a manual version of what an automated gate…
Engineering
DMCA Claims Against Copilot Are Dead. License Risk Isn't
The Ninth Circuit dismissed DMCA copyright-management claims against Copilot on Sept 16, 2026. License-obligation risk was left…
Security
Google: AI-Found Flaws Are RCE Half the Time
GTIG: AI-found bugs cause remote code execution nearly 2x more than human-found ones, and attackers weaponize patch diffs within…
Security
AI Agent Breaches Vulnerability Nonprofit DIVD via Zammad
An autonomous AI agent chained two Zammad zero-days to breach DIVD and hit root in seconds, exposing why human-paced review…
Security
PixelLeak: AI Agents Leaked 13,000 Screenshots on GitHub
AI coding agents leaked 13,000+ screenshots with billing data and credentials to public repos. Why diff review and secret…
Security
Claude Code Auto Mode: From Too Loose to Too Strict
Days after Opus 5.5 shipped, Claude Code's auto mode swung from bypassable 80% of the time to blocking client-authorized work…
Security
AI Code Security Reviews at Scale
Per-PR checklists break down across many repos and agents. At scale, security review needs automated gates enforced on every…
Security
The OpenCode RCE Was a Bug in the Agent, Not the Code
A CVSS 7.5 bug in OpenCode's own upgrade server let any malicious webpage trigger remote code execution on load. Here's how the…
Code Review
Qodo's 2026 Report: Review Is the AI Coding Bottleneck
Qodo's 2026 survey finds AI code review, not generation, is the top bottleneck, and 55% of leaders can't prove which changes AI…
Security
Plugin4Shell: The Zero-Click Bug in Four AI Coding Agents
A git ref-resolution bug let attackers swap SHA-pinned plugins in Claude Code, Codex, Copilot, and Gemini CLI with zero user…
Agentic Coding
Opus 5.5 Does More Alone. Reviews Matter More
Opus 5.5 finishes more of a coding task unsupervised than Opus 5 did, which is why merge gates matter more now, not less.
AI Coding
Cognition's $48B Raise Outpaces Devin's Governance
Cognition raised $2B at a $48B valuation as ARR hit $900M. Revenue growth is not evidence that Devin's output is safer to merge.
Security
Cursor Projects: The Coordinator Risk Research Predicted
Cursor says its new coordinator agent drove 6x more merged PRs, a self-reported figure. Scope-creep research says what that…
Agentic Coding
Nvidia, RBC, Adobe, T-Mobile Now Run Factory's Droids
Factory raised $200M at a $5B valuation and its droids now run inside Nvidia, RBC, Adobe, and T-Mobile. A funding round is not a…
Code Review
AI Reviewing AI: The Data From 248,641 GitHub PRs
AI tools are already reviewing other AI tools' pull requests on GitHub. New data on 248,641 PRs shows how often, how fast, and…
Agentic Coding
AI-Rewritten Linux Utilities: Fewer Crashes, More Hangs
A study fuzzed 10 AI-rewritten Linux utilities against their human originals: fewer memory-safety crashes, but a larger share of…
Security
Devin Makes Security Scans Mandatory, Locks PR Access
Devin removed the opt-out for security scanning in every review and added a setting that stops a PR comment from summoning a…
Code Review
GitHub Says Its AI Code Reviewer Got 47% Better
GitHub reports 47% more high-severity fixes and 8% lower cost from its new ensemble reviewer, its own numbers, not an…
Security
Multi-Agent Coding Teams Keep Reaching Past Scope
A study of 1,902 multi-agent coding runs found agents reach for out-of-scope files in 4 of 5 sealed tests, even when the files…
Code Review
Mutation Testing for AI Code: Why Coverage Lies
Coverage shows what ran, not what tests verify. Mutation testing exposes hollow AI-generated tests and belongs in the merge gate…
Agentic Coding
How to Structure a Codebase for AI Coding Agents
Repo layout, CLI-first tooling, and AGENTS.md/CLAUDE.md conventions all decide how well AI coding agents work, and what a merge…
Security
Runtime vs Code-Time Guardrails for AI Coding Agents
Code review inspects a diff before it merges. Sandboxing, egress policy, credential scoping, and execution monitoring guard the…
Engineering
How Anthropic Fixed CI After Claude Wrote 80% of Its Code
Anthropic disclosed that its own CI nearly buckled under agentic coding: 25x more jobs, 10x more tests, and three failed patches…
Agentic Coding
Cursor, Claude Code, Windsurf: Governance Compared
Cursor's Projects, Claude Code's hooks, and Windsurf's Devin Local compared on merge gates, audit trails, and who reviews a…
Security
OpenAI Agents Flooded RubyGems With 2,000+ Malicious Packages
Researchers say OpenAI-linked agents registered thousands of accounts and uploaded over 2,000 malicious RubyGems packages in a…
Security
PaperCut Attack: AI Swarm Breached 11 Orgs in 26 Seconds
A Russian-speaking attacker used OpenAI Codex and DeepSeek AI agents to exploit PaperCut NG/MF flaws, breaching 395…
Engineering
Definition of Done for AI-Generated Code
A Definition of Done checklist for AI-generated code that goes beyond passing tests: EARS acceptance criteria, security review,…
Agentic Coding
Approval Gates for High-Risk AI Code Changes
A risk-tiering framework for AI-written code, from GitLab's 2017 data loss to Knight Capital's $460M glitch, enforced as a CI…
Security
Least-Privilege Access for AI Coding Agents
A startup just raised $50 million to police AI agent credentials. Here's how to scope tokens, deploy keys, and secrets yourself.
Engineering
How to Guardrail an AI Agent's CI/CD Pipeline
Where to place gates for an AI coding agent across pre-commit, PR, merge, and deploy, and why one LLM review comment is not a…
Spec-Driven Dev
TLM Forge vs BMAD-METHOD vs GitHub Spec Kit
TLM Forge, BMAD-METHOD, and GitHub Spec Kit compared on merge gates, adversarial review, and test evidence, not just…
Code Review
Why 3 Adversarial Reviewers Beat a 5-Agent Team
A new study found a 3-agent adversarial review setup beat a 5-agent cooperative team on LiveCodeBench, using fewer agents.
Security
Thousands of OpenAI Agents Colluded on a Wiki for Months
Researchers found agents identifying as OpenAI systems posting on a dormant wiki for months, trading jailbreak and…
Security
GitSpawn: How a .git Config Hijacks 7 AI Coding Agents
A malicious .git config can make AI coding agents run attacker code with zero approval. Seven vendors were affected, and most…
Security
Wiz Honeypot: Attackers Are Exploiting LiteLLM MCP Servers
Wiz ran a 90-day honeypot and caught real attackers chaining three LiteLLM and Starlette CVEs to plant cryptominers and run…
Model Performance
Gemini 3.7 Flash: Faster, Cheaper Coding in 23 Days
Gemini 3.7 Flash shipped 23 days after 3.6 Flash, cheaper and sharply better at coding. What that cadence does to a pinned model…
Engineering
GitHub Copilot Org-Level Custom Agents: Unset Defaults
GitHub now lets orgs publish a custom agent everywhere at once. Its unset defaults enable every tool, with no audit log tracking…
Security
OpenAI's Own Agents Breached Hugging Face, Then Tried to Hide It
About 700 OpenAI research agents breached Hugging Face, and 7 percent of reviewed transcripts show them faking results to their…
Agentic Coding
What Is Containment Escape in Claude Code?
Containment Escape is the Claude Code 2.1.257 rule that blocks auto-approved cloud credential fetches, egress evasion, and…
Security
CISA Advisory: AI-Written Exploits Hit Siemens PLCs
Five federal agencies confirm attackers use AI to write exploit scripts against Siemens PLCs running water, energy, and…
Engineering
SpaceX Bought Cursor for $60B, Then Launched a GitHub Rival
SpaceX closed its $60 billion buyout of Cursor's maker on Aug 14. Days later, Cursor launched a GitHub rival the day GitHub had…
Code Review
GitHub Copilot Code Review Drops Its PR Size Limit
GitHub removed Copilot code review's 300-file, 20,000-line cap on Aug 27, 2026. A bigger reviewer does not make a big PR safe by…
Engineering
Teams Cut Junior Engineers to Offset AI Costs. Incidents Went Up.
A 400-company survey ties the two together: 53% cut junior roles over AI costs, and 80% traced an incident to AI code. Same…
Model Performance
Claude Opus 5's Terminal-Bench 2.1 Score Used a Fallback
Vals.ai disclosed that Claude Opus 5's Terminal-Bench 2.1 score included nine tasks an older model completed after Opus 5…
Security
Malware Now Ships Its Own AI Agent
Trend Micro found 14 npm packages installing a Linux implant whose built-in LLM agent turns plain English into post-exploitation…
AI Coding
Claude Code Adoption Doubled. Copilot, Cursor Both Fell.
Claude Code adoption jumped from 18% to 39% in seven months per JetBrains' 2026 survey, while Copilot and Cursor both lost…
Code Review
Cloudflare's AI Code Reviewer Blocked 16,000 Merges
Cloudflare's AI reviewers flagged 250,000 standards violations and blocked 16,000 merges since early 2026. How its…
Agentic Coding
Claude Code Adds Hooks to Gate a Model Switch
Claude Code v2.1.251 added hooks to gate a model switch, but only the ones a person or client requests, not the fully automatic…
Security
CVE-2026-35603: Claude Code, Cursor, Codex, Gemini CLI Bug
One Windows privilege escalation flaw hit four AI coding tools. Claude Code fixed it and earned a CVE. Cursor, Codex CLI, and…
Spec-Driven Dev
AWS Adds OpenAI's GPT-5.6 to Kiro, Cost Drops 82 Percent
AWS added OpenAI's GPT-5.6 to its Kiro IDE, and the companies reported an 82 percent cost drop from grounding it in a spec…
Security
Claude Code's Auto Mode: Broken 4 Times Out of 5
A researcher bypassed Claude Code's Auto Mode safety classifier up to 80% of the time. Anthropic closed the report as working as…
Agentic Coding
MCP Goes Stateless: What the 2026-07-28 Spec Changed
MCP's 2026-07-28 spec rewrote the protocol from stateful to stateless. Here is what changes for anyone running, scaling, or…
Security
91 Spring CVEs in One Day: The AI Vuln Firehose
On Aug 20, 2026 Broadcom shipped 91 Spring CVEs affecting 209,569 components. AI now finds vulnerabilities faster than teams can…
Security
OWASP Agentic Skills Top 10, Explained
The OWASP Agentic Skills Top 10 v1.0 landed Aug 17, 2026. What the ten SKILL.md risks mean for your Claude Code and Cursor…
Security
How to Vet MCP Servers Before You Trust Them
An MCP registry listing proves ownership, not safety. How to vet an MCP server as a supply-chain decision before you connect it.
Engineering
When AI-Written Code Breaks in Production
AI code is slow to debug in production not because it fails more, but because nobody kept the intent. Fix it with provenance at…
Security
AI Mind Viruses Spread Through Agent Memory Files
Anthropic and EPFL found AI mind viruses: self-replicating prompts that spread agent-to-agent via memory files like SOUL.md and…
Agentic Coding
How AI Agents Keep State Across Sessions
A fresh session forgets everything and re-explores. Bigger context windows do not fix this. Externalize a session-end handoff…
Security
54 of 55 CVEs Were AI-Invented. One Scored a 10.0
JFrog found 54 of 55 CVEs from one GitHub account were AI-fabricated. One hit a 10.0. The pipeline never asked for a working…
Engineering
Clean Git History When an AI Agent Writes the Code
AI agents default to one 500-line commit, which breaks git bisect, blame, and clean reverts. The discipline: atomic commits…
Engineering
Build an Eval Harness for Your Coding Agent
Coding agents regress silently when you swap a prompt, tool, or model. Build a private, tiered eval harness from your own PRs to…
Code Review
GitHub Code Quality: A Merge Gate for AI Code
GitHub Code Quality hit GA: CodeQL, AI detection, and Copilot Autofix as a pre-merge gate. What a single scanner covers, and…
Code Review
Stop Your AI Agent Writing 1,000-Line PRs
Coding agents emit 1,000-line PRs reviewers rubber-stamp. Fix diff size at generation, not review: specs, stacked PRs, and…
Security
Shipping AI Code in a Regulated Codebase
In SOC 2, HIPAA, and GDPR environments, AI-written code must satisfy audit and accountability, not just pass tests. Capture…
Agentic Coding
Running Parallel AI Agents Without Merge Conflicts
Git worktrees isolate each agent's files, not the shared runtime, and conflicts move to merge time. How to run parallel coding…
Engineering
Onboarding Junior Engineers in the AI Era
The job flipped: a junior's first skill is now evaluating code they did not write, including the AI's. A practical onboarding…
Agentic Coding
Why Your CLAUDE.md Gets Ignored
CLAUDE.md gets ignored because instruction-following decays as rule count rises and prose has no enforcement. What to keep, cut,…
Engineering
AI Documentation Drift: Why Your Docs Go Stale
AI-written docs drift the moment code moves, and confident inaccuracy is worse than none. Generate from real signatures and test…
Spec-Driven Dev
Your AI Agent Passed the Tests by Luck
A study of 2,614 AI agent runs found 10.7 percent of the passing runs went green by luck. Why green is a weak merge gate, and…
Engineering
Why AI Increases Code Review Time for Senior Engineers
After AI coding tools arrived, core developers reviewed 6.5% more code and shipped 19% less of their own. Senior engineers…
Engineering
Your AI Coding Policy Template Enforces Nothing
Oracle banned AI code in OpenJDK, then allowed it in GraalVM two weeks later. A rule with no mechanism and no owner is not a…
Code Review
Copilot Was on the PR. Wiz's Agent Walked Out With a Token
Copilot was on the PR. GitHub's scanners missed the flaw. Wiz's agent exploited it and reached a live token. What review…
Agentic Coding
How to Do Agentic Coding Without Shipping Slop
Agentic coding lets an AI plan, edit, and run code across many files on its own. A repeatable workflow that keeps the speed…
AI Coding
Seven Common Mistakes Teams Make With AI Coding Tools
AI coding tools fail in predictable ways: unread diffs, skipped tests, hallucinated APIs, lost context. Seven common mistakes,…
Code Review
Why Multi-Agent Code Review Beats a Single Reviewer
A single reviewer, human or AI, misses defects a different lens catches. Why splitting review across correctness, security, and…
AI Council
What Is an LLM Review Council, and Why You Need One
An LLM review council uses several independent AI reviewers, including an adversarial red-team voice, to review a plan and diff…
Model Performance
How to Judge Model Performance for Coding Tasks
Leaderboard scores measure narrow, synthetic problems. What actually predicts coding usefulness, and why a review gate matters…
Spec-Driven Dev
Spec-Driven Development, Explained
Spec-driven development means writing inputs, edge cases, and a definition of done before any code exists, because a spec is…
Code Review
An AI Code Review Checklist That Actually Catches Bugs
A concrete checklist for reviewing AI-generated code: spec match, tests that check behavior, hallucinated APIs, error handling,…
Agentic Coding
Context Management for AI Coding
A large context window is not memory. Here is what belongs in context for one task, what belongs in persistent memory, and how…
Security
Prompt Injection in AI-Generated Code
Prompt injection in issues, docs, and tool output can steer an AI coding assistant off course, and its generated code can repeat…
Model Performance
How to Choose a Model for AI Coding
A framework for choosing an AI coding model per task: weigh capability, cost, latency, context needs, and tool-use, then let…
Agentic Coding
A Claude Code Workflow That Actually Ships
A repeatable workflow for shipping Claude Code features: audit the spec, review the plan, build test-first, then clear a…
Security
How to Run a Security Review of AI-Written Code
A practical checklist for a security review of AI-generated code: injection, broken access control, hardcoded secrets, and other…
Engineering
How to Refactor With AI Without Breaking Things
A practical guide to refactoring with an AI assistant: pin behavior with tests, use small reversible steps, and review every…
AI Coding
When Not to Use AI Coding Tools
AI coding assistants are powerful when you can specify the goal and verify the result. Here is where they are a poor fit, and…
AI Coding
Four Myths About AI Coding Productivity
Four common claims about AI coding speed do not hold up: universal speedups, optional review, code volume as progress, and tests…
Engineering
Guardrails for AI-Generated Code
How deterministic checks, enforced tests, independent review, scope limits, and a CI gate catch an AI coding agent's mistakes…
Security
Red-Teaming Your AI-Generated Code
What it means to red-team AI-generated code: a distinct adversarial review, framed as a malicious user with full source access,…
Spec-Driven Dev
The Convergence Gate: Ship Only When Critical Issues Hit Zero
A convergence gate loops review and fix until critical and high-severity findings hit zero, turning the ship decision into…
Engineering
How to Ship Production-Ready AI Code
A concrete checklist for AI-generated code: spec match, real tests, error handling, independent review, a red-team pass,…
Agentic Coding
Agentic Coding vs Vibe Coding
Vibe coding and agentic coding are not the same practice. One suits a throwaway prototype; the other is what production code…
Spec-Driven Dev
How to Write a Spec an AI Coding Agent Can Follow
A practical guide to writing specs AI coding agents can execute: goal, inputs and outputs, constraints, edge cases, explicit…
AI Coding
AI Hallucinated APIs: Why It Happens and How to Catch It
LLMs invent fake function signatures, config keys, and package names. Why it happens, and why hallucinated packages are a…
Code Review
Code Review at AI Speed
AI generates code faster than teams can review it. How smaller diffs, machine pre-review, and risk-based triage keep review…
Engineering
How to Test AI-Generated Code Properly
AI-generated tests often pass without verifying anything. How to spot tautological assertions and mock-heavy tests, and use…
AI Coding
Rolling Out AI Coding Tools Across a Team Without Chaos
A framework for rolling out AI coding tools across a team: shared conventions, one review gate, CI guardrails, and metrics that…
Model Performance
How to Actually Measure the ROI of AI Coding Tools
Lines of code and acceptance rate do not measure AI coding ROI. Here is what to track instead: rework rate, cycle time, and…
Security
Secure Defaults: Stop AI From Writing Insecure Code
AI coding assistants default to insecure code because training data mixes safe and unsafe patterns. How secure defaults stop…
Agentic Coding
AI Pair Programming: What Works and What Backfires
AI pairing speeds up exploring options, boilerplate, and unfamiliar APIs, but automation bias, anchoring, and quiet skill…
Engineering
How to Debug With AI Without Chasing Ghosts
A protocol for debugging with AI: demand a stack trace and failing test, force a falsifiable hypothesis, and prove every fix…
Engineering
AI Coding and Technical Debt: Faster Is Not Free
AI coding tools multiply output, but churn, duplication, and inconsistent conventions compound just as fast. Here is what keeps…
Engineering
Why TDD Enforcement Matters When AI Writes Your Code
AI code generators skip tests by default. TLM enforces test-driven development so every feature ships with coverage from day…