The claude_local adapter runs Anthropic’s Claude Code CLI locally. It supports session persistence, skills injection, and structured output parsing.

​
Prerequisites

  • Claude Code CLI installed (claude command available)
  • ANTHROPIC_API_KEY set in the environment or agent config

​
Configuration Fields

FieldTypeRequiredDescription
cwdstringYesWorking directory for the agent process (absolute path; created automatically if missing when permissions allow)
modelstringNoClaude model to use (see Supported models below — defaults to the Claude Code CLI’s default when unset)
effortstringNoReasoning effort. Valid: low, medium, high, max. Maps to claude --effort. Use low for cheap subagents, high (default) for complex work, max for frontier problems.
fallbackModelstringNoAlternative Claude model ID used when the primary model is overloaded. Maps to claude --fallback-model. Typical pairing: primary=claude-opus-4-7, fallback=claude-sonnet-4-6.
maxBudgetUsdnumberNoHard USD cap for a single run. Maps to claude --max-budget-usd. The CLI aborts the run once the cap is reached. Complements FideliOS’s per-agent monthly budgets.
betasstringNoComma- or space-separated Anthropic beta-header names forwarded via claude --betas. API-key users only. See Beta features below.
promptTemplatestringNoPrompt used for all runs
envobjectNoEnvironment variables (supports secret refs)
timeoutSecnumberNoProcess timeout (0 = no timeout)
graceSecnumberNoGrace period before force-kill
maxTurnsPerRunnumberNoMax agentic turns per heartbeat (defaults to 300)
dangerouslySkipPermissionsbooleanNoSkip permission prompts (dev only)

​
Supported models

The adapter’s model list mirrors whatever the installed Claude Code CLI exposes. As of FideliOS 0.0.29 the dropdown shows:
Model IDLabel
claude-opus-4-7Claude Opus 4.7
claude-opus-4-6Claude Opus 4.6
claude-sonnet-4-6Claude Sonnet 4.6
claude-haiku-4-6Claude Haiku 4.6
Leave model unset to defer to the Claude Code CLI’s default (recommended for most agents). Explicit overrides are useful when you want to pin a specific model for cost or capability reasons.

​
Beta features

The Claude Code CLI (v2.1.101+) forwards Anthropic beta headers via the --betas flag. Set the betas field on your agent’s adapter config to opt in. These are API-key-only features — they do not work with subscription auth.

​
Advisor tool — Opus 4.7 as strategic advisor for cheaper executors

model: claude-sonnet-4-6
betas: "advisor-tool-2026-03-01"
Lets a faster, cheaper executor (Sonnet or Haiku) consult Opus 4.7 mid-generation for strategic advice. Typical saving: 30-50% of cost at near-Opus-solo quality on agentic tasks. Works well on long-horizon coding / research pipelines where the plan matters more than every output token. See Anthropic’s Advisor Tool docs.

​
Task budgets — Opus 4.7 self-pacing

model: claude-opus-4-7
betas: "task-budgets-2026-03-13"
maxBudgetUsd: 2.50
Gives Opus 4.7 a token countdown so it paces itself over a long agentic loop and finishes gracefully before the limit. Complements maxBudgetUsd (which is a hard cap — task-budget is a soft hint). Opus 4.7 only. See Task Budgets docs.

​
Fast mode — 2.5x output speed on Opus 4.6

model: claude-opus-4-6
betas: "fast-mode-2026-02-01"
Up to 2.5x faster output tokens per second. 6x the token price. Waitlist-gated, Opus 4.6 only. Useful for real-time tasks; not a default. See Fast Mode docs.

​
Combining betas

Space or comma separate to pass multiple: 
betas: "advisor-tool-2026-03-01, task-budgets-2026-03-13"

​
Prompt Templates

Templates support {{variable}} substitution:
VariableValue
{{agentId}}Agent’s ID
{{companyId}}Company ID
{{runId}}Current run ID
{{agent.name}}Agent’s name
{{company.name}}Company name

​
Session Persistence

The adapter persists Claude Code session IDs between heartbeats. On the next wake, it resumes the existing conversation so the agent retains full context. Session resume is cwd-aware: if the agent’s working directory changed since the last run, a fresh session starts instead. If resume fails with an unknown session error, the adapter automatically retries with a fresh session.

​
Skills Injection

The adapter creates a temporary directory with symlinks to FideliOS skills and passes it via --add-dir. This makes skills discoverable without polluting the agent’s working directory. For manual local CLI usage outside heartbeat runs (for example running as claudecoder directly), use: 
pnpm fidelios agent local-cli claudecoder --company-id <company-id>
This installs FideliOS skills in ~/.claude/skills, creates an agent API key, and prints shell exports to run as that agent.

​
Environment Test

Use the “Test Environment” button in the UI to validate the adapter config. It checks:
  • Claude CLI is installed and accessible
  • Working directory is absolute and available (auto-created if missing and permitted)
  • API key/auth mode hints (ANTHROPIC_API_KEY vs subscription login)
  • A live hello probe (claude --print - --output-format stream-json --verbose with prompt Respond with hello.) to verify CLI readiness