Tasks

Nightshift includes 59 built-in tasks organized by category. See the Task Reference for a complete list.

Browse Tasks

# List all tasks
nightshift task list

# Filter by category
nightshift task list --category pr
nightshift task list --category analysis

# Filter by cost tier
nightshift task list --cost low
nightshift task list --cost medium

# Show task details
nightshift task show lint-fix
nightshift task show lint-fix --prompt-only

Categories

Category	Description
pr	Creates PRs with code changes
analysis	Produces analysis reports without code changes
options	Suggests improvements for human review
safe	Execution or simulation with no lasting side effects
map	Codebase mapping and documentation
emergency	Preparedness artifacts for incident response

Cost Tiers

Tier	Token Usage	Examples
low	Minimal	lint-fix, docs-backfill
medium	Moderate	dead-code, test-gap
high	Significant	bug-finder, auto-dry
veryhigh	Large	migration-rehearsal, contract-fuzzer

Run a Single Task

# Dry run first
nightshift task run lint-fix --provider claude --dry-run

# Execute
nightshift task run lint-fix --provider claude

Skill Grooming Task

Nightshift includes a built-in skill-groom task for keeping project-local skills aligned with the current codebase.

It is enabled by default. You can set its interval and priority:

tasks:
  priorities:
    skill-groom: 2
  intervals:
    skill-groom: "168h"

To opt out:

tasks:
  disabled:
    - skill-groom

skill-groom uses README.md as project context, checks .claude/skills and .codex/skills, and validates SKILL.md content against the Agent Skills format (starting docs lookup from https://agentskills.io/llms.txt).

Custom Tasks

Define custom tasks in your config:

tasks:
  custom:
    - type: pr-review
      name: "PR Review Session"
      description: |
        Review all open PRs. Fix obvious issues immediately.
        Create tasks for bigger problems.
      category: pr
      cost_tier: high
      risk_level: medium
      interval: "72h"

Custom tasks use the same scoring, cooldowns, and budget controls as built-in tasks. The description field becomes the agent prompt. Only type, name, and description are required — other fields have sensible defaults.

Task Cooldowns

Each task has a default cooldown per project. After running lint-fix on ~/code/sidecar, it won't run again on that project for 24 hours. Override with tasks.intervals in config.

Category defaults:

Category	Default Interval
PR	7 days
Analysis	3 days
Options	7 days
Safe	14 days
Map	7 days
Emergency	30 days

Use nightshift preview --explain to see cooldown status, including which tasks are currently on cooldown and when they become eligible again. When all tasks for a project are on cooldown, the run is skipped.

td Review Task

The td-review task runs a detailed review session over open td reviews. It:

• Reviews all open review tasks in the project
• Fixes obvious bugs immediately and creates td bug tasks for them
• Creates new td tasks with detailed descriptions for bigger issues
• Verifies changes have tests, creates tasks for missing coverage
• Uses subagents for reviews that can run in parallel
• Closes in-progress tasks once related bugs are resolved

This task is disabled by default and must be explicitly opted in:

tasks:
  enabled:
    - td-review

Requires the td integration to be enabled (see Integrations).