configure

This skill should be used when the user asks to "configure the agentic workflow", "set up imat-agentic", "run the setup wizard", "change the confidence threshold", "update workflow settings", or when

SKILL.md

GitHub

Skill: configure

Interactively creates or updates .claude/imat-agentic.local.md — the project-local configuration for the imat-agentic plugin.

When this runs

Run this before the first agentic-loop invocation, or when settings need to change. This file is gitignored (.claude/*.local.md) — each contributor configures their own preferences.

Configuration fields

Key Default Options Description
quality_preset sonnet sonnet, quality, opus Model tier for most agents
parallel_explorers 2 1, 2, 3 Concurrent exploration instances
confidence_threshold 80 70, 80, 90 Review filter threshold
max_review_retries 2 1–5 Max review retry cycles
harness_profile full full, light, off Harness check scope
recall_before_explore true true, false Run recall before exploration
auto_advance false true, false Skip confirmations (use for automation only)
artifact_retention_days 30 integer Days before cleaning up old agentic/ artifacts

Wizard procedure

1. Check for existing config

Read .claude/imat-agentic.local.md if it exists. Show current values before asking for changes.

2. Ask each question

Use AskUserQuestion for each configurable field. Show the default and options clearly. Allow accepting defaults by selecting the first option.

3. Write the config file

Write .claude/imat-agentic.local.md:

---
quality_preset: sonnet
parallel_explorers: 2
confidence_threshold: 80
max_review_retries: 2
harness_profile: full
recall_before_explore: true
auto_advance: false
artifact_retention_days: 30
---

# imat-agentic configuration

Local configuration for the imat-agentic plugin.
This file is gitignored — do not commit it.

Run `/imat-agentic:configure` to update these settings.

4. Confirm

Show a summary of the written config. Remind the user that the file is gitignored and each contributor sets their own.

Notes

  • auto_advance: true disables user confirmation prompts mid-loop. Only use for scripted/CI runs.
  • quality_preset: opus uses the most capable model but costs more per run.
  • confidence_threshold: 70 surfaces more review findings — useful when onboarding to a new codebase. Set to 90 once the codebase is well-understood.