Tag: claude-code

A claude -p run on Opus with an API key gives up after three consecutive 529s, about 1.5 seconds of backoff, no matter what CLAUDE_CODE_MAX_RETRIES says. Add --fallback-model, set CLAUDE_CODE_RETRY_WATCHDOG=1 with a max wait, or widen the 529 backoff with the OVERLOADED_RETRY delay variables. Checked against Claude Code 2.1.309.

Claude Code 2.1.289 and earlier resend every screenshot on every turn, so one refused request means every later turn is refused too. Update to 2.1.290+ and resume with claude --continue; the client then trims the oldest images to 510, 255, then 127. On an old build, /compact or /clear. Then stop the screenshots piling up.

Since Claude Code v2.1.295, a command or HTTP hook can set onFailure to "block" so a guard that crashes, times out, cannot start or prints invalid JSON stops the tool call instead of waving it through. Here is the config, what counts as a failure, the events where it does nothing, and the exit-0 trap no setting can catch.

Claude Code v2.1.296 lets a subagent auto-compact earlier than the main conversation. Set autoCompactWindow in the subagent's frontmatter or --agents JSON as a plain integer from 100000 to 1000000, and watch out for the 200k parsing trap, the CLAUDE_CODE_AUTO_COMPACT_WINDOW override, and the --agent session mode where the field does nothing.

GitHub Agentic Workflows (gh-aw v0.91.5) compiles a Markdown file into a hardened GitHub Actions .lock.yml that runs Copilot, Claude Code, Codex, Gemini or Pi in a read-only, firewalled job and applies writes through safe-outputs. What it adds on top of Actions, the jobs it generates, and when a plain workflow with claude-code-action is still the better tool.

A context window is everything the model can see in one request: system prompt, tool definitions, history, tool results, thinking and the reply. Here is what counts, what happens at the limit on claude-opus-5-5 and friends, and the six techniques agents use to work past it: deferred tools, context editing, compaction, memory, subagents and output filtering.

An agent skill is a folder with a SKILL.md file that the agent loads on demand: only its name and description sit in the system prompt, the instructions arrive when a task matches, and bundled scripts run without ever entering context. Here is how that differs from a system prompt in cost, scope, invocation and trust, with the exact limits from the Agent Skills spec and Claude Code.

MCP is a JSON-RPC protocol that lets any AI host call tools, read resources and use prompts from any server, the way LSP did for language support. Here is how it works on the wire under revision 2026-07-28, captured from a real server, and why VS Code, Visual Studio, JetBrains, Xcode, Zed and Cursor all converged on it.

A coding agent refactors a component, renames a data-testid, and your Playwright suite goes red. Here is a three-layer setup, verified on Playwright 1.63.0 and Claude Code 2.1.197: locators that survive, a PreToolUse hook that blocks the rename, and a CI guard that lets the Playwright healer agent fix locators but never rewrite an assertion.

Every agent request pays for its system prompt, the provider's hidden tool-use preamble, and every tool schema before the user says anything. How to measure that fixed cost with differential count_tokens calls on the Anthropic API, the input_tokens endpoint on OpenAI, an MCP listTools probe, and /context in Claude Code, anchored to claude-opus-5-5, claude-sonnet-5, and @modelcontextprotocol/client 2.1.0.