SET is the thin orchestration layer in the ABVX stack.

It does not replace coding agents. It makes repo workflow choices explicit, keeps outputs predictable for CI and downstream agent runs, and coordinates repo context, optional human context, and review artifacts through one small entrypoint.

• orchestrates agentsgen repo-doc and pack flows;
• integrates optional ID human-context hooks;
• exports review-first planning artifacts before any workflow write;
• keeps registry-driven repo presets and capability contracts in one place.

• it is not an agent runtime;
• it is not a prompt library;
• it does not replace agentsgen, ID, or delivery-time proof/release gates;
• it does not write target repos directly in planning mode.

Use the action in a target repository:

- uses: markoblogo/SET@main
  with:
    workflow_preset: repo-docs
    path: "."

Review-first planning pass:

python3 scripts/plan_config_apply.py markoblogo/<owner>/<repo> --format json

Useful local planning commands:

python3 scripts/plan_config_apply.py markoblogo/lab.abvx
python3 scripts/plan_config_apply.py markoblogo/lab.abvx --dry-run --format json
python3 scripts/plan_config_apply.py markoblogo/lab.abvx --export-dir ./.set-plan
python3 scripts/plan_config_apply.py markoblogo/lab.abvx --repo-root /path/to/lab
python3 scripts/validate_registry.py

SET is intentionally review-first:

resolve repo config -> plan -> inspect -> approve -> apply in target repo

Owner-facing and agent-facing phase vocabulary stays compact:

Discuss -> Plan -> Execute -> Verify -> Ship

In SET, heavy research, planning, and execution work should run in bounded fresh packets while the main thread stays lean. A ship claim is valid only after a verification artifact or receipt exists for the reviewed scope.

The planner can export:

• plan.json
• workflow.set.yml
• pr-body.md
• orchestrator-bundle.json
• proposal-lifecycle.json
• optional rabbithole.seed.md

See docs/config-apply-planning.md and docs/orchestrator-compatibility.md.

• repo-docs: init + pack + check
• site-ai: repo-docs + site pack + analyze + meta
• minimal: bootstrap-only baseline

When repo-local ID hooks are enabled, SET resolves the preferred human bootstrap order and exports:

• docs/ai/id-bootstrap.json
• docs/ai/id-bootstrap.prompt.md

Expected bootstrap order:

1. soul.md
2. profile.core.md
3. handshake.md

SET treats those packets as runtime interfaces for downstream agent runs, not as hidden action internals. See docs/id-bootstrap.md.

Depending on preset and config, SET can drive production of:

• AGENTS.md, RUNBOOK.md
• llms.txt / LLMS.md
• docs/ai/id-context.json
• docs/ai/id-bootstrap.json
• docs/ai/id-bootstrap.prompt.md
• repo maps and AI docs under docs/ai/
• proof-loop artifacts under docs/ai/tasks/<task-id>/

When execution docs or operator-facing runbooks are meant to be acted on under high context-switching or ADHD-style working conditions, prefer ADHD-shaped output: next action first, numbered bounded steps, visible current state, concrete estimates, and one explicit next move.

• agentsgen: repo-scoped agent context
• ID: portable human context
• lab.abvx: public catalog and read-only surface

Practical split:

• use ID for the human;
• use agentsgen for the repository;
• use SET when you want orchestration around both.

SET owns the orchestration registry and repo-config contract:

• schema: schema/repo-config.v1.json
• docs: docs/repo-config.md
• runtime artifact docs: docs/id-bootstrap.md
• example: examples/repo-config.example.json
• registry entries: registry/repos/*.json

SET can export compact optional contracts without becoming a runtime for them.

Core orchestration and proof:

• docs/loop-readiness-contract.md
• docs/loop-hardening-contract.md
• docs/bounded-orchestration-contract.md
• docs/ship-router-contract.md
• docs/bug-evidence-contract.md
• docs/agent-governance-capability-contract.md
• docs/task-closeout-audit-contract.md

Context, memory, and agent operations:

• docs/memory-capability-contract.md
• docs/memory-mutation-receipt-contract.md
• docs/agent-operations-contract.md
• docs/skill-quality-pipeline-contract.md
• docs/gepa-style-optimization-contract.md
• docs/codex-security-repo-selection-matrix.md
• docs/search-discoverable-code-contract.md
• docs/twelve-factor-agent-adaptation.md

Frontend, product, and content review:

• docs/creator-intent-lens.md
• docs/design-taste-review-contract.md
• docs/design-blueprint-contract.md
• docs/readme-surface-audit-contract.md
• docs/dependency-risk-check-contract.md
• docs/chat-surface-boundary-contract.md
• docs/code-review-helper-boundary-contract.md

Operational references and analytics:

• docs/agent-traversal-review-note.md
• docs/terminal-agent-multiplexer-reference.md
• docs/external-ai-engineering-references.md
• docs/ai-visibility-audit-contract.md
• docs/donor-adaptation-target-matrix.md
• docs/500-ai-agents-donor-shortlist.md
• docs/project-portfolio-adaptation-matrix.md
• docs/agents-marketplace-owner-routing.md

For profile selection and export behavior, see docs/capability-profiles.md.

Start here:

• docs/repo-config.md
• docs/config-apply-planning.md
• docs/orchestrator-compatibility.md
• docs/id-bootstrap.md

Then read by need:

• context budget and diversity: docs/context-budget-hint.md, docs/research-diversity-hint.md
• scope and capability routing: docs/v0.1-scope.md, docs/capability-profiles.md, docs/llmo-capability-map.md
• reference material: docs/references/open-notebook.md, docs/references/agent-orchestrators.md, docs/donor-adaptation-target-matrix.md, docs/project-portfolio-adaptation-matrix.md, docs/agents-marketplace-owner-routing.md

The top-level README stays focused on:

• what SET is;
• how to run it;
• what it emits;
• where deeper contracts live.

Detailed adaptation contracts remain in docs/ so the homepage does not turn into an always-growing changelog of every imported idea.

See CONTRIBUTING.md.