Notes from the background.
RSS →Deep dives on running Claude Code unattended — the CLI, output formats, scheduling, and the background agent patterns that start paying off around ticket #4.
- 6 min
Claude Code background commands: Ctrl+B, auto-cleanup, and the 5GB limit
Claude Code can run bash commands in the background — dev servers, build watchers, test runners. A field guide to Ctrl+B, the run_in_background parameter, output retrieval via Read, the Ctrl+X Ctrl+K kill switch, and the 5GB cap nobody mentions.
Read the full piece → - 7 min
What is 'Background Claude'? A field guide to the term
Background Claude is the umbrella term for running Claude Code without a human watching it — headless, scheduled, or agent-triggered. The short definition, the three levels, the four properties that set it apart, and where the phrase came from.
Read the full piece → - 8 min
Linear CLI + claude -p: the poor-man's background agent in 30 lines of shell
The Linear CLI plus claude -p plus a while loop gives you a real Linear-triggered background agent in 30 lines of bash. A field guide to the loop, and the exact four things it breaks on.
Read the full piece → - 9 min
Claude Code pricing: what it actually costs to run background agents
Claude Code charges by API token consumption — not per seat. Average cost is $13/developer/active day. The real pricing model, the cost caps every background agent needs, and six strategies to keep spend predictable.
Read the full piece → - 6 min
Turn claude -p into a multi-turn REPL with one named pipe
claude -p is designed for one-shot prompts. --input-format stream-json plus a FIFO plus the fd-3 trick turns it into an interactive multi-turn REPL you drive from shell commands. Eight lines of bash, line-by-line.
Read the full piece → - 8 min
Claude Code in GitHub Actions: the v1 recipe, with guardrails
anthropics/claude-code-action@v1 is GA. Three workflows worth the CI minutes (@claude mentions, auto PR review, scheduled automation), the minimal viable YAML, and the beta→v1 migration that broke every existing recipe.
Read the full piece → - 9 min
Claude Code permission modes: which one should your CI use?
There are six Claude Code permission modes — default, acceptEdits, plan, auto, dontAsk, and bypassPermissions. They are not interchangeable. A field guide to which fits which CI scenario, grounded in the official docs.
Read the full piece → - 7 min
git worktree for Claude Code: the isolation pattern every background agent uses
Two concurrent Claude Code runs on the same repo will fight over the git index. git worktree gives each run its own checkout, its own branch, its own install — zero races, zero locks, zero mystery. The pattern every background agent ends up implementing.
Read the full piece → - 6 min
Claude Code --bare: the flag the docs want to be the default
--bare skips hooks, skills, plugins, MCP servers, CLAUDE.md auto-discovery, OAuth, and the keychain. The docs say it'll become the default for claude -p. Here's what it turns off and why CI wants exactly that.
Read the full piece → - 8 min
Claude Code stream-json: the output format that changes everything
stream-json is the third --output-format value and nobody talks about it. It's newline-delimited JSON events, it turns the CLI into a live dashboard, and it's what makes background agents possible. A field guide.
Read the full piece →