Configuration reference

Every adamast.json field, default, and precedence rule is defined on this page; it is the canonical reference. Other pages show only the fields they need.

📥 How loading works

• AdaMAST reads one dependency-free JSON file, adamast.json by default, or the path given by --config.
• Explicit CLI/API arguments win over config-file values.
• Unknown top-level fields are rejected so spelling mistakes fail loudly. Keys inside the claude_code and codex blocks are read individually; a misspelled host key is ignored rather than rejected.
• Relative paths are resolved relative to the config file; ~ is expanded.
• "version" must be 1 (or omitted).

Credentials

AdaMAST never stores credential values. Set provider keys in your environment.

🪶 Minimal config

Most projects only need this:

{
  "version": 1,
  "trace_output": "./adamast-program",
  "adamast_model": "gpt-5"
}

Everything else has a sensible default.

🧱 Core fields

Field	Default	Purpose
trace_output	required	Program-specific folder for traces, evidence, and manifest state. This is the program identity: same folder = same program.
adamast_model	required for learning	Model used by AdaMAST generation, judging, and refinement. Keep it separate from your task-solving model.
trace_root	~/.adamast/traces	Root for per-taxonomy trace folders after acceptance.
store_dir	~/.adamast/taxonomies	Flat taxonomy store: one JSON file per taxonomy_id.
inherit	unset	taxonomy_id to start from. Unset (or the string "none") starts from built-in MAST.
dashboard	true	Let integrations launch the localhost dashboard automatically.
evidence_export	unset	Optional external snapshot sink for session-end evidence. A .json value is written as that exact file; any other value is treated as a directory receiving one <program_id>.json per program.
redact_traces	true	First-party adapters strip credential-looking substrings (API keys, tokens, cookies) from traces before persistence. Disable only when traces are known clean and exact text matters.

Environment overrides, read when the corresponding value is not set explicitly: ADAMAST_HOME moves the ~/.adamast base directory, ADAMAST_TRACE_ROOT and ADAMAST_STORE_DIR override the two storage roots, and ADAMAST_DISABLE_DASHBOARD=1 stops integrations from launching the dashboard.

📈 Learning fields

Field	Default	Purpose
generation_threshold	5	Warm-up trace count before generating a taxonomy from MAST. Raise it if early traces are noisy or unrepresentative.
generation_stops	false	If true, the current task waits for generation to finish; if false, running tasks continue on MAST and the result activates afterwards.
skip_judge	false	Skip Reflection-Judge cleanup of generated taxonomies (structural validation still runs).
k_init	10	Traces required before the first refinement of an active stored taxonomy.
k	20	Traces required between later refinements for the same program.
refinement_stops	false	If true, the current task waits for refinement to finish.
advanced_refinement	false	Add one support-judge repair pass during refinement.
freeze	false	Inference-only mode: record traces and evidence but skip generation and refinement. Use for pinned-taxonomy A/B evaluations.

Hook hosts always learn in the background

Hook integrations (Claude Code, Codex) always run generation and refinement in background workers regardless of generation_stops / refinement_stops: a hook process is killed at the harness's per-hook timeout, so inline learning is honored only by the single-LLM and CLI paths that own their process.

See TRACES_AND_LEARNING.md for how these interact.

🚦 Gate fields

These fields budget the runtime checkpoints (the reflection boundaries also called gates) and the final pre-submission checkpoint. Form failures (unparseable reflections) and substantive repairs draw on separate budgets, so tuning one loop cannot silently disable the other.

Field	Default	Purpose
repair_rounds	3	Substantive REPAIR_REQUIRED repair opportunities at the final gate before honest unresolved release.
format_retries	2	Reflection parse/shape failures tolerated per checkpoint cycle before the gate releases. Format re-prompts are targeted: they name the missing elements and pin the previous verdict.
max_retries	3	Legacy shared knob, accepted indefinitely: maps to repair_rounds when repair_rounds is not set.
gate_exhaustion_policy	"raise"	Single-LLM only. "raise" errors when the repair cap is hit; "release" returns the best answer and records gate_allowed=false (useful for benchmark wrappers).

🔌 Model transport fields

Transport is selected by model-id shape; these fields adjust it:

Field	Default	Purpose
model	unset	Task-solving model for adamast single-run (same as its --model flag). Not used by learning calls.
openai_base_url	unset	OpenAI-compatible endpoint override; also honored via the OPENAI_BASE_URL environment variable.
openai_api_key_env	unset	Name of the environment variable holding the key for that endpoint. Store the variable name, never the value.

Credential expectations per provider are covered in INSTALLATION.md.

🧩 Integration-scoped fields

Field	Default	Purpose
project_dir	unset	Project directory for the install CLIs (same as --project-dir).
claude_code	unset	Claude Code adapter policy. claude_code.built_in_hooks enables/disables built-in events and sets tool matchers; see CLAUDE_CODE.md.
codex	unset	Codex adapter policy under codex.hooks; see CODEX.md.
failure_throttle_calls	5	Claude Code: minimum tool calls between repeated PostToolUseFailure nudges for the same failure.
failure_recency_seconds	30	Claude Code: minimum seconds between repeated failure nudges.
recent_activity_messages	8	Single LLM: trajectory-window size (messages) delivered at checkpoints.
recent_activity_chars	12000	Single LLM: trajectory-window size (characters) delivered at checkpoints.

The legacy top-level built_in_hooks, custom_hooks, and codex_hooks fields are still accepted as compatibility aliases for the scoped forms.

User-level installs use different defaults

The defaults in the tables below describe explicit project installs. adamast codex install --user-level instead defaults to project_scope: "auto", session_selector: "prompt", selector_surface: "browser", and learning_backend: "codex_subagent". adamast claude install --user-level uses the parallel Claude values with learning_backend: "claude_subagent". Both user-level commands default to the shared root ~/.adamast/interactive and runtime identity interactive-session.

Codex user-level interactive hooks may also set:

Field	Default	Purpose
codex.project_scope	"explicit"	"auto" derives a separate program from each canonical project root.
codex.project_id	unset	Optional stable identity override, useful for intentionally joined worktrees.
codex.task_group	"default"	Project-local trace, taxonomy, and refinement group.
codex.session_selector	"off"	Set to "prompt" to ask for MAST, a compatible stored taxonomy, or AdaMAST-off at the start of a new Codex conversation.
codex.selector_surface	"browser"	"browser" opens the session-bound local library from the first UserPromptSubmit; "inline" presents the same choices in chat. SessionStart never launches a new browser selector.
codex.learning_backend	"provider"	Set to "codex_subagent" for durable learning through a native subagent in the active Codex task, without a separate API key or CLI login.
codex.worker_model	unset	Legacy compatibility field; native spawned subagents use the active Codex task's model policy.
codex.codex_cli_path	unset	Legacy compatibility field; not required by native in-task learning.
codex.worker_timeout_seconds	1800	Claim lease for each native generation/refinement and support-review subagent.

Claude Code accepts parallel interactive fields:

Field	Default	Purpose
claude_code.project_scope	"explicit"	"auto" derives a program from the canonical project root.
claude_code.project_id	unset	Optional stable project identity override.
claude_code.task_group	"default"	Logical project-local taxonomy and refinement group. Automatic routing namespaces it for Claude Code, separately from Codex.
claude_code.session_selector	"off"	"prompt" asks for MAST, a compatible taxonomy, or AdaMAST-off.
claude_code.selector_surface	"inline"	"browser" opens the session-bound local library. Both install commands write "browser" at every level, so "inline" applies only when a hand-written config omits the key.
claude_code.learning_backend	"provider"	"claude_subagent" uses native generator and support-review Agent subtasks without a separate API key or CLI login.
claude_code.worker_model	unset	Legacy compatibility field; the native Agent follows the active session's model policy.
claude_code.claude_cli_path	unset	Legacy detached-worker compatibility field; native in-session learning does not use it.
claude_code.worker_timeout_seconds	1800	Browser-selection wait limit and claim lease for each native generation/refinement and support-review Agent. The installed browser UserPromptSubmit hook adds a 15-second shutdown margin.
claude_code.custom_hooks	[]	Extra AdaMAST checkpoint hooks. Each entry: required name and event, optional mode ("blocking"/"advisory"), matcher, command_pattern, and checkpoint_key ("tool_use_id"/"command"/"fixed"). Managed by adamast claude add-hook / remove-hook / list-hooks; see Claude Code.

🏷️ Display metadata

Field	Default	Purpose
repo	auto-discovered	Display-only repository label shown in the picker and dashboard. Never routes taxonomy selection.
repo_path	unset	Directory used to auto-discover the display repo label from git metadata.

✅ Full example

{
  "version": 1,
  "trace_output": "./adamast-program",
  "trace_root": "~/.adamast/traces",
  "store_dir": "~/.adamast/taxonomies",
  "adamast_model": "gpt-5",
  "inherit": null,
  "generation_threshold": 5,
  "generation_stops": false,
  "skip_judge": false,
  "k_init": 10,
  "k": 20,
  "refinement_stops": false,
  "advanced_refinement": false,
  "freeze": false,
  "repair_rounds": 3,
  "format_retries": 2,
  "dashboard": true,
  "claude_code": {
    "built_in_hooks": {
      "SubagentStop": false,
      "PostToolUse": ["Bash", "Edit"],
      "PostToolUseFailure": {
        "enabled": true,
        "matchers": ["Bash"]
      }
    }
  }
}

Validate any config with:

adamast doctor --config adamast.json

➡️ Continue with

• Traces and learning: how the learning fields interact at runtime.
• Troubleshooting: when a resolved value does not behave as expected.