How szarkans/multi works
multi is an AI-agent plugin and shell/Python runtime that fans a task out to configured model backends, then presents their independent opinions. Its principal workflow is code review: select a profile, capture an immutable code snapshot, run reviewers in parallel (with a separate overengineering lens), and consolidate corroborated, disputed, checked, or dropped findings. The README also documents question answering, completion checks, and ADHD planning. Source excerpts are limited, so some runtime relationships below reflect README statements rather than inspected implementations.
Main entry points: ask.sh Task dispatch · providers.sh Backend runners · _parser.py TOML parser · SKILL.md Code review
Components 20
Agent entrypoints
- Code review
skills/code-review/SKILL.md - Ask
skills/ask/SKILL.md - Completion check
skills/check-if-done/SKILL.md - ADHD planning
skills/adhd/SKILL.md - Setup guidance
skills/setup/SKILL.md
Task runtime
- Task dispatch
scripts/ask.sh - Code snapshot
scripts/snapshot.sh - Review prompts
scripts/review-prompt.sh - Backend runners
scripts/providers.sh - OpenCode report
scripts/opencode-report.py - Idea parsing
scripts/parse-branch.py - Run artifacts
scripts/run-dir.sh
Backend configuration
- Profile resolution
scripts/config.py
TOML support
- TOML API
scripts/_vendor/tomli/__init__.py - TOML parser
scripts/_vendor/tomli/_parser.py - Value parsing
scripts/_vendor/tomli/_re.py - Parser types
scripts/_vendor/tomli/_types.py
Other components
- Developer
- AI backends
- Overengineering lens
Connections 22
- Developer → Code review: requests review
README.md - Developer → Ask: asks question
README.md - Developer → Completion check: checks completion
README.md - Developer → ADHD planning: requests planning
README.md - Code review → Task dispatch: invokes task
- Ask → Task dispatch: invokes task
- Completion check → Task dispatch: invokes task
- ADHD planning → Task dispatch: invokes task
- Task dispatch → Profile resolution: resolves profile
scripts/config.py - Profile resolution → TOML API: parses config
scripts/config.py - TOML API → TOML parser: exports parser
scripts/_vendor/tomli/__init__.py - TOML parser → Value parsing: uses patterns
scripts/_vendor/tomli/_parser.py - TOML parser → Parser types: uses types
scripts/_vendor/tomli/_parser.py - Value parsing → Parser types: uses types
scripts/_vendor/tomli/_re.py - Task dispatch → Code snapshot: captures snapshot
README.md - Task dispatch → Review prompts: prepares review
- Task dispatch → Backend runners: runs in parallel
README.md - Backend runners → AI backends: queries models
README.md - Task dispatch → Overengineering lens: runs separate lens
README.md - Backend runners → OpenCode report: formats output
- Task dispatch → Idea parsing: parses planning output
- Task dispatch → Run artifacts: stores run artifacts