Configuration Reference
Three layers: four .env process vars, ~45 database-backed runtime keys on the Config page, and per-issue overrides in IssueConfig. All runtime config takes effect on the next daemon poll cycle - no restart needed.
SQLite backed
Runtime editable
Per-issue overrides
Multi-harness
Environment Variables
Read once at process start from .env or the shell. Cannot change without restarting the server.
| Variable | Type | Default | Description |
|---|---|---|---|
DATABASE_URL |
string | ~/.slop/data/slop.db |
Path to the SQLite database file. Override only when you need the DB outside the default location. |
PORT |
number | 3000 |
Next.js HTTP port. Convention for local dev is 3100. |
SLOP_SECRET |
string | (unset) | Optional bearer token gating the web UI. When unset, server binds to 127.0.0.1 only. When set, all requests must carry Authorization: Bearer <SLOP_SECRET>. |
SLOP_URL |
string | http://localhost:3100 |
Base URL that harness skills use when calling back into Slop. Must be reachable from wherever the agent subprocess runs. |
GITHUB_TOKEN: Provide this in the environment on first boot and the daemon seeds it into the
githubToken Config key automatically. After that it is read from the database and the env var is no longer required.
Watched Repo
Watched repos are Repo rows added on the Config page. The keys below are the global connection settings plus the legacy single-repo keys (used during the one-shot migration to the multi-repo model).
| Key | Type | Default | Description |
|---|---|---|---|
githubToken |
string | (must set) | Personal access token (or fine-grained token) with repo and workflow scopes. Used for all GitHub API calls across all watched repos. Seeded from GITHUB_TOKEN env var on first boot. |
baseBranch |
string | main |
Default base branch for PRs and fast-forward merges. Can be overridden per Repo row on the Config page. |
watchedRepo |
string | (none) | Legacy single-repo owner/name slug. Consumed once during migration to the Repo table. New installs add repos on the Config page instead. |
watchedRepoPath |
string | (none) | Legacy absolute path to the local checkout. Consumed once during migration. New installs set path on the Config page. |
Automation Modes
Each flag is stored as "true" or "false" in the Config table. They compose independently: you can enable auto-merge without auto-mode, or auto-review without auto-address.
autoMode
Default: false
Master auto-pilot. When true, the daemon claims issues from the ready queue each poll cycle (up to parallelismCap). When false, issues must be dispatched manually from the board.
autoMergeMode
Default: true
When true, Slop enables GitHub auto-merge on the PR after the agent pushes. The PR merges automatically once all required CI checks pass. When false, a human must merge manually.
autoReportMode
Default: false
When true, the daemon automatically generates and posts a completion report after each worker finishes, without waiting for a manual trigger.
autoReviewMode
Default: false
When true, the daemon automatically dispatches a /pr-review skill session for PRs entering waiting_review status. Works with autoAddressMode for a fully autonomous review cycle.
autoAddressMode
Default: false
When true, the daemon automatically dispatches an /address-comments session after a /pr-review verdict that does not meet acceptedReviewLevel. Requires autoReviewMode to also be true.
Address pipeline dependency:
autoAddressMode is gated on autoReviewMode. Enabling address-mode without review-mode has no effect - a review must run first to produce comments to address.
Implementation Settings
Implement Mode
remote
Remote Mode
Agent commits, pushes branch, opens a PR. Slop enables GitHub auto-merge, waits for CI checks to pass, then merges via the GitHub API. Full PR review pipeline is available in this mode.
local (default)
Local Mode
Agent commits to a local branch. Slop fast-forward merges directly into the base branch - no PR is opened and no GitHub merge API call is made. Faster, but no CI gate or PR review pipeline.
| Key | Type | Default | Description |
|---|---|---|---|
implementMode |
enum | local |
How the agent ships work. remote or local. See cards above. |
verifyGate |
boolean | false |
When true, a /verify-gate skill session runs after implementation. Worker advances only on SLOP_VERDICT: pass. Any other or missing verdict counts as findings and re-dispatches implement. |
agentModel |
enum | opus |
Model used for implementation. Determines the harness (claude / codex / copilot). See model catalog below. |
agentEffort |
enum | medium |
SDK effort level. Values: low, medium, high, xhigh, max. Claude harness only: ultracode maps to max and opts into multi-agent Workflow orchestration. |
agentTimeoutMs |
number | 1800000 (30 min) |
Hard timeout per individual agent session/query. If exceeded, the session is killed and the worker fails or retries. |
implementTimeoutMs |
number | 3600000 (60 min) |
Total wall-clock timeout for the entire implement phase. |
verifyTimeoutMs |
number | 3600000 (60 min) |
Timeout for the verify-gate skill session. |
verifyCommand |
string | (none) | Custom shell command the verify agent runs to check the implementation. When unset, the harness uses its default verification skill. |
pointsHarness |
enum | claude |
Which harness tier map to use for story-point scoring: claude, codex, or copilot. Unrecognized values fall back to claude. |
Agent Model Catalog
| Model value | Label | Harness | Provider |
|---|---|---|---|
opus | Claude Opus | claude | Anthropic |
sonnet | Claude Sonnet | claude | Anthropic |
haiku | Claude Haiku | claude | Anthropic |
fable | Claude Fable | claude | Anthropic |
gpt-5.5 | GPT-5.5 | codex | OpenAI |
gpt-5.4 | GPT-5.4 | codex | OpenAI |
copilot-claude-sonnet-4.6 | Claude Sonnet 4.6 | copilot | GitHub |
copilot-claude-sonnet-4.5 | Claude Sonnet 4.5 | copilot | GitHub |
copilot-claude-haiku-4.5 | Claude Haiku 4.5 | copilot | GitHub |
copilot-gpt-5.4 | GPT-5.4 | copilot | GitHub |
copilot-gpt-5.3-codex | GPT-5.3-Codex | copilot | GitHub |
copilot-gpt-5-mini | GPT-5 mini | copilot | GitHub |
The model value is the single source of truth for which harness handles a worker - the harness is derived from the model prefix, not stored separately.
Concurrency and Polling
| Key | Type | Default | Description |
|---|---|---|---|
parallelismCap |
number | 1 |
Maximum number of workers that can be actively implementing at the same time. Workers in non-implementing statuses (waiting_ci, merging, etc.) do not count against this cap. Per-repo overrides are set on individual Repo rows. |
pollIntervalMs |
number | 30000 (30 s) |
How often the daemon poll cycle runs. Read fresh from the Config table each cycle, so changes take effect without a restart. Shorter intervals react faster to CI status changes but increase GitHub API usage. |
pollPaused |
boolean | false |
When true, the daemon skips the refreshRepoSnapshot step each cycle and stops claiming new issues. Useful during maintenance. Existing workers continue to completion. |
Attempt Budgets
When a counter hits its cap, the worker transitions to failed. A Retry button appears on the board to reset counters and re-queue.
maxCiAttempts
5
CI fix cycles. Each
fixing_ci agent session that ends in failure consumes one attempt.maxConflictAttempts
5
Merge conflict resolution attempts. Each
resolving_conflict session consumes one.maxVerificationAttempts
5
Verify-gate cycles. Each
verifying session that does not return pass consumes one.maxReviewAttempts
3
PR review-and-address cycles. Each
in_review session consumes one.maxResumeAttempts
3
Times a crashed or orphaned
implementing worker can auto-resume against its saved session after a daemon restart.Stuck Detection Thresholds
When a worker stays in a given status longer than the configured threshold, the daemon sends a notification via notificationTarget. Workers are not automatically failed - the alert lets a human decide whether to retry or cancel. All values are in milliseconds.
Important: Keys not seeded at startup (
stuckFixingCiMs, stuckVerifyingMs, stuckWaitingReviewMs, stuckInReviewMs, stuckWaitingAddressMs, stuckInAddressMs) fall back to hardcoded defaults in lifecycle-poll.ts when absent from the Config table. Setting them explicitly overrides the hardcoded value.
| Key | Default (ms) | Human | Status watched |
|---|---|---|---|
stuckClaimedMs |
300000 |
5 min | claimed - worker created but agent never started. Short because claiming should be near-instant. |
stuckWaitingCiMs |
3600000 |
60 min | waiting_ci - PR open, waiting for CI checks to complete. |
stuckMergingMs |
1800000 |
30 min | merging - auto-merge enabled, waiting for GitHub to merge. May indicate auto-merge is blocked. |
stuckResolvingConflictMs |
1800000 |
30 min | resolving_conflict - agent is resolving a merge conflict. |
stuckFixingCiMs |
1800000 |
30 min | fixing_ci - agent is fixing CI failures. |
stuckVerifyingMs |
1800000 |
30 min | verifying - verify-gate skill session is running. |
stuckWaitingReviewMs |
3600000 |
60 min | waiting_review - PR is open and awaiting a human or automated review. |
stuckInReviewMs |
1800000 |
30 min | in_review - /pr-review skill session is running. |
stuckWaitingAddressMs |
3600000 |
60 min | waiting_address - waiting for comments to be addressed. |
stuckInAddressMs |
1800000 |
30 min | in_address - agent is actively addressing review comments. |
stuckRenotifyMs |
86400000 |
24 hr | Re-notification cadence. How long to wait before re-alerting about the same stuck worker. Not a status threshold. |
PR Review Settings
| Key | Type | Default | Description |
|---|---|---|---|
acceptedReviewLevel |
enum | non_blockers |
The minimum /pr-review verdict that counts as approved for advancing to merge. non_blockers: only non-blocking findings (strictest autonomous pipeline). nitpick: only nitpick-level or better. none: any verdict passes including blockers. |
Per-Issue Overrides (IssueConfig)
Each issue can have individual overrides stored in the IssueConfig table, keyed by (repoId, issueNumber). Overrides are copied onto the Worker row when the issue is claimed and take precedence over global config. null means "fall back to global."
Two surfaces:
IssueConfig overrides are set before claiming (from the issue card on the board). Worker overrides are set on an active worker via the gear popover in the worker list and take effect on the next agent dispatch.
| Field | Type | Description |
|---|---|---|
agentModelOverride |
enum? | Override agentModel for this issue. Must be a value from the model catalog. Determines the harness. |
agentEffortOverride |
enum? | Override agentEffort for this issue. |
implementModeOverride |
enum? | Override implementMode (remote or local) for this issue only. |
autoMergeModeOverride |
string? | Override autoMergeMode for this issue. Three-way: null = use global, "true" = enable, "false" = disable. |
autoReviewModeOverride |
string? | Override autoReviewMode for this issue. Same three-way semantics. |
autoAddressModeOverride |
string? | Override autoAddressMode for this issue. Same three-way semantics. |
acceptedReviewLevelOverride |
enum? | Override acceptedReviewLevel for this issue only. |
storyPoints |
number? | Story point estimate (1-5 scale). Used for automatic model/effort selection when modelEffortFromStoryPoints is true. |
pointsAutoScored |
boolean | Default false. Set to true when story points were assigned automatically by the scoring agent rather than set manually. |
modelEffortFromStoryPoints |
boolean | Default true. When true and storyPoints is set, the configured harness tier map picks model and effort automatically. When false, explicit overrides or global config are used instead. |
Miscellaneous
| Key | Type | Default | Description |
|---|---|---|---|
notificationTarget |
string | "" |
Notification destination (e.g. a webhook URL or identifier). When empty, all notifications are suppressed. |
memoryThresholdBytes |
number | 2147483648 (2 GiB) |
RSS memory threshold for active agent subprocesses. When a running process exceeds this, memoryBreachAction is triggered. |
memoryBreachAction |
enum | warn |
What to do when memoryThresholdBytes is exceeded. warn: log and notify only. stop: stop the worker after reporting. |
usdToBrlRate |
number | (none) | Exchange rate for converting USD agent costs to BRL in the board UI. When unset, only USD costs are shown. |
costCoefficient |
number | (none) | Multiplier applied to raw API cost figures before display. Useful to account for markup or seat fees. When unset, raw cost is shown. |
guardStatus |
string | (internal) | Setup wizard state cached by the daemon. Managed internally - do not set manually. |
baseCiStatus |
string | (internal) | Base branch CI health cached by the CI health controller. Goes red after 2 consecutive failing cycles (anti-flake debounce). New workers are not claimed when red. |
Full Config Key Reference
Alphabetical index of every CONFIG_KEYS constant and which section covers it.
| Key | Type | Default | Section |
|---|---|---|---|
acceptedReviewLevel | enum | non_blockers | PR Review Settings |
agentEffort | enum | medium | Implementation Settings |
agentModel | enum | opus | Implementation Settings |
agentTimeoutMs | number | 1800000 | Implementation Settings |
autoAddressMode | boolean | false | Automation Modes |
autoMergeMode | boolean | true | Automation Modes |
autoMode | boolean | false | Automation Modes |
autoReportMode | boolean | false | Automation Modes |
autoReviewMode | boolean | false | Automation Modes |
baseBranch | string | main | Watched Repo |
baseCiStatus | string | (internal) | Miscellaneous |
costCoefficient | number | (none) | Miscellaneous |
githubToken | string | (must set) | Watched Repo |
guardStatus | string | (internal) | Miscellaneous |
implementMode | enum | local | Implementation Settings |
implementTimeoutMs | number | 3600000 | Implementation Settings |
maxCiAttempts | number | 5 | Attempt Budgets |
maxConflictAttempts | number | 5 | Attempt Budgets |
maxResumeAttempts | number | 3 | Attempt Budgets |
maxReviewAttempts | number | 3 | Attempt Budgets |
maxVerificationAttempts | number | 5 | Attempt Budgets |
memoryBreachAction | enum | warn | Miscellaneous |
memoryThresholdBytes | number | 2147483648 | Miscellaneous |
notificationTarget | string | "" | Miscellaneous |
parallelismCap | number | 1 | Concurrency and Polling |
pointsHarness | enum | claude | Implementation Settings |
pollIntervalMs | number | 30000 | Concurrency and Polling |
pollPaused | boolean | false | Concurrency and Polling |
stuckClaimedMs | number | 300000 | Stuck Detection |
stuckFixingCiMs | number | 1800000 | Stuck Detection |
stuckInAddressMs | number | 1800000 | Stuck Detection |
stuckInReviewMs | number | 1800000 | Stuck Detection |
stuckMergingMs | number | 1800000 | Stuck Detection |
stuckRenotifyMs | number | 86400000 | Stuck Detection |
stuckResolvingConflictMs | number | 1800000 | Stuck Detection |
stuckVerifyingMs | number | 1800000 | Stuck Detection |
stuckWaitingAddressMs | number | 3600000 | Stuck Detection |
stuckWaitingCiMs | number | 3600000 | Stuck Detection |
stuckWaitingReviewMs | number | 3600000 | Stuck Detection |
usdToBrlRate | number | (none) | Miscellaneous |
verifyCommand | string | (none) | Implementation Settings |
verifyGate | boolean | false | Implementation Settings |
verifyTimeoutMs | number | 3600000 | Implementation Settings |
watchedRepo | string | (legacy) | Watched Repo |
watchedRepoPath | string | (legacy) | Watched Repo |