Config reference

Location: ~/.sybra/config.yaml. Loaded once at startup. A small set of environment variables override selected fields at runtime — see Environment variables.

Complete schema

# ─────────────────────────────────────────────────────────────
# Logging
# ─────────────────────────────────────────────────────────────
logging:
  level: "info"                   # debug | info | warn | error
  dir: "~/.sybra/logs"
  max_size_mb: 50
  max_files: 5

# ─────────────────────────────────────────────────────────────
# Audit
# ─────────────────────────────────────────────────────────────
audit:
  enabled: true
  retention_days: 30

# ─────────────────────────────────────────────────────────────
# Notifications
# ─────────────────────────────────────────────────────────────
notification:
  desktop: true

# ─────────────────────────────────────────────────────────────
# Agent defaults (overridable per task / per workflow step)
# ─────────────────────────────────────────────────────────────
agent:
  provider: "claude"              # claude | codex
  model: "sonnet"                 # opus | sonnet | haiku | full model ID
  mode: "headless"                # headless | interactive
  max_concurrent: 3
  max_cost_usd: 5.0
  max_turns: 150
  require_permissions: true       # null = true, false = --dangerously-skip-permissions
  max_log_events: 500             # events returned when replaying log
  research_machine_dir: ""

# ─────────────────────────────────────────────────────────────
# Orchestrator
# ─────────────────────────────────────────────────────────────
orchestrator:
  auto_triage: true
  auto_plan: true

# ─────────────────────────────────────────────────────────────
# Integrations
# ─────────────────────────────────────────────────────────────
todoist:
  enabled: false
  api_token: ""
  project_id: ""
  default_project_id: ""
  poll_seconds: 120

github:
  enabled: true

renovate:
  enabled: true
  author: "app/renovate"

# ─────────────────────────────────────────────────────────────
# Background workers
# ─────────────────────────────────────────────────────────────
triage:
  enabled: false
  poll_seconds: 60
  model: "sonnet"

monitor:
  enabled: true
  interval_seconds: 300
  model: "sonnet"
  issue_cooldown_minutes: 30
  dispatch_limit: 3
  stuck_human_hours: 8.0
  lost_agent_minutes: 15
  failure_rate_threshold: 0.3
  bottleneck_hours:
    plan-review: 4.0
    human-required: 8.0
    in-progress: 6.0
    default: 12.0
  issue_label: "monitor"
  issue_repo: "Automaat/sybra"

self_monitor:
  enabled: false
  interval_hours: 6.0
  judge_model: "claude-haiku-4-5-20251001"
  synthesizer_model: "claude-sonnet-4-6"
  max_issues_per_run: 5
  max_auto_actions_per_day: 3
  auto_act_categories:
    - stuck_task
    - workflow_loop
    - cost_outlier
    - triage_mismatch
  dry_run: true
  issue_cooldown_hours: 24.0
  issue_label: "selfmonitor"
  max_cost_per_tick_usd: 2.0
  judge_parallelism: 4
  suppression_days: 7
  suppression_threshold: 3

# ─────────────────────────────────────────────────────────────
# Providers
# ─────────────────────────────────────────────────────────────
providers:
  health_check:
    enabled: true
    interval_seconds: 300
  claude:
    enabled: true
    rate_limit_cooldown_seconds: 900
  codex:
    enabled: true
    rate_limit_cooldown_seconds: 900
  auto_failover: true

# ─────────────────────────────────────────────────────────────
# Metrics
# ─────────────────────────────────────────────────────────────
metrics:
  enabled: false

# ─────────────────────────────────────────────────────────────
# Per-machine routing
# ─────────────────────────────────────────────────────────────
project_types: []                 # empty = all; ["work", "pet"] = only these

# ─────────────────────────────────────────────────────────────
# Directory overrides
# ─────────────────────────────────────────────────────────────
tasks_dir: "~/.sybra/tasks"
skills_dir: "~/.claude/skills"
projects_dir: "~/.sybra/projects"
clones_dir: "~/.sybra/clones"
worktrees_dir: "~/.sybra/worktrees"
loop_agents_dir: "~/.sybra/loop-agents"

Field reference

logging

Field	Type	Notes
level	string	debug, info, warn, error
dir	path	Directory for rotating logs
max_size_mb	int	Per-file rotation threshold
max_files	int	Files kept after rotation

audit

Field	Type	Notes
enabled	bool	Writes ~/.sybra/logs/audit/*.ndjson
retention_days	int	Files older than this are deleted

agent

Field	Type	Notes
provider	string	claude or codex
model	string	Short alias or full model ID
mode	string	headless or interactive
max_concurrent	int	Global cap on simultaneous agents
max_cost_usd	float	Per-run cost ceiling. Escalates on hit.
max_turns	int	Per-run turn ceiling. Escalates on hit.
require_permissions	bool	true pauses on every tool_use. false = skip-permissions.
max_log_events	int	Events replayed from log when opening agent detail

orchestrator

Field	Type	Notes
auto_triage	bool	Let orchestrator classify new tasks
auto_plan	bool	Auto-plan medium/large tasks

todoist

Field	Type	Notes
enabled	bool	Feature kill switch
api_token	string	Personal API token. Override at runtime with SYBRA_TODOIST_TOKEN.
project_id	string	Fetch from this Todoist project
default_project_id	string	Target when creating tasks in Todoist
poll_seconds	int	Poll interval

github

Field	Type	Notes
enabled	bool	Feature kill switch

Credentials come from local gh auth — no token in config.

renovate

Field	Type	Notes
enabled	bool	Feature kill switch
author	string	GitHub username/login that authors Renovate PRs

triage

Field	Type	Notes
enabled	bool	Auto-triage poller (independent of orchestrator)
poll_seconds	int	Poll interval
model	string	Model for the triage classifier

monitor

Field	Type	Notes
enabled	bool	In-process anomaly detector
interval_seconds	int	Poll interval
model	string	Model used for remediation decisions
issue_cooldown_minutes	int	Same issue won’t be re-posted within this window
dispatch_limit	int	Max remediation agents per cycle
stuck_human_hours	float	Flag human-required tasks older than this
lost_agent_minutes	int	Agent running past this with no heartbeat = “lost”
failure_rate_threshold	float	Flag rolling failure rate above this (0.0–1.0)
bottleneck_hours	map	Per-status hours in column before flagging
issue_label	string	Label applied to meta issues
issue_repo	string	Repo where meta issues are posted

self_monitor

Field	Type	Notes
enabled	bool	Hourly retrospective judge
interval_hours	float	Tick interval
judge_model	string	Fast model scoring runs
synthesizer_model	string	Strong model consolidating findings
max_issues_per_run	int	Ceiling on findings per tick
max_auto_actions_per_day	int	Ceiling on auto-remediations per day
auto_act_categories	list	Finding types eligible for auto-action
dry_run	bool	true = report only, never act
suppression_days	int	Suppress duplicates within this window
suppression_threshold	int	Seen this many times = suppress
max_cost_per_tick_usd	float	Budget cap per tick
judge_parallelism	int	Concurrent judge runs per tick

providers

Field	Type	Notes
health_check.enabled	bool	Periodic Claude/Codex probe
health_check.interval_seconds	int	Probe interval
claude.enabled / codex.enabled	bool	Individual provider kill switch
claude.rate_limit_cooldown_seconds	int	After 429, wait this long before retrying
auto_failover	bool	On provider fail, fall back to the other

project_types

Empty = accept all project types on this machine. Non-empty = only tasks tied to projects with matching type run here. See Orchestrator and automation → Per-machine routing.

Directory overrides

Paths support ~ expansion. Use to point Sybra at a shared NFS mount for multi-machine setups (route carefully via project_types).

---

Validation

• Unknown field → warning in log, ignored
• Wrong type → Sybra falls back to last-known-good config
• Missing required (schema-level) → startup fails with a clear error

On startup, Sybra logs an app.automations summary with enabled toggles. Check it to verify the role of each machine.

---

Reset

Delete ~/.sybra/config.yaml. Sybra writes a fresh default on next start.