Prompt injection: why there is no definitive fix, and what to do anyway
Twelve prompt injection defenses reporting near-zero rates were broken above 90%. Why the problem is structural and what reduces risk.
- prompt injection
- agent security
Page 2 of 4
All posts
Git with an agent: commit, branch and assisted code review
The list of git operations auto mode refuses on its own, worktrees for parallel work, and why /rewind does not undo what a background review wrote.
- claude code
- git
Plugins and marketplaces: how to evaluate before installing
The documentation says a plugin runs arbitrary code with your privileges. The five-minute checklist using the tools already shipping in Claude Code.
- claude code
- plugins
Skills vs subagents vs commands: which to use for what
One of the three stopped being a separate category. The four questions that pick the right mechanism, and the precedence table that flips order between them.
- claude code
- skills
What each Claude Code usage pattern costs
48,000 real turns measured across 67 sessions. Where the money goes, why turn 400 costs three times turn 10, and how much compaction gives back.
- claude code
- cost
The 18 Claude Code traps in a real project
Eighteen documented behaviors that degrade in silence. The list is organized by symptom: what you see before you know what caused it.
- claude code
- traps
Context: why your agent forgets and how to fix it
Three mechanisms make Claude Code look forgetful, and each has its own fix. What compaction keeps, what it drops, and what costs you.
- claude code
- context
Permissions and auto mode: what to clear and what to lock
Deny beats ask, which beats allow, and specificity changes nothing. Claude Code's three permission layers and the rules that look protective without protecting.
- claude code
- permissions
Claude Code, Codex and Cursor: what actually differs
All three adopted the same open standards and the feature comparison ended in a tie. What still differs is where context lives and who enforces restrictions.
- claude code
- codex
Slash commands: turning a process into a command
A custom command became a skill. What changes: positional arguments, context injected before the agent sees it, and stacking up to six in one message.
- claude code
- slash commands
MCP in practice: connecting Claude Code
The three installation scopes, the precedence that does not merge fields, and the trap of MCP's local scope, which differs from settings' local.
- claude code
- mcp
Motion •
prefers-reduced-motion: accessible motion without killing the design
The spec's word is reduce, not remove. Why animation: none !important gets it wrong on both sides, and the substitution table that preserves the design.
- accessibility
- motion design