Thin orchestration repo for the ABVX development tools ecosystem.
It stays deliberately small: presets define a baseline, and explicit inputs override the preset when you need per-repo control.
Current working definition:
agentsgen= repo intelligence runtimeSET= orchestration layer / GitHub Action entrypointlab.abvx= public catalog and read-only control plane- standalone tools such as
git-tweetstay independent and are integrated by contract
This repo stays intentionally thin even after adding registry, planning, and drift-check layers.
- uses: markoblogo/SET@main
with:
workflow_preset: "site-ai"
agentsgen: "true"
init: "true"
pack: "true"
site_pack: "true"
site_url: "https://example.com"
check: "true"
repomap: "true"
repomap_compact_budget: "4000"
repomap_focus: "cli"
repomap_changed: "false"
analyze: "true"
analyze_url: "https://example.com"
meta: "true"
meta_url: "https://example.com"
proof_loop: "true"
proof_task_id: "proof-loop-v0"
autodetect: "true"
path: "."Preset baselines:
minimal-> repo bootstrap onlyrepo-docs-> init + pack + checksite-ai-> repo-docs + site pack + analyze + meta
What v0.1 does:
- installs
agentsgen - runs
agentsgen init - optionally runs
agentsgen pack - optionally runs
agentsgen pack --site <url> - optionally runs
agentsgen check --all --ci - optionally runs
agentsgen understand --compact-budget <tokens> - optionally runs
agentsgen snippets - optionally runs
agentsgen analyze <url> - optionally runs
agentsgen meta <url> - optionally runs
agentsgen task init/evidence/verdictas a proof-loop hook, including richer evidence/verdict summaries - supports
workflow_presetbaselines with explicit input override - writes a compact GitHub Actions summary for the resolved run plan
- passes a first-class repomap policy through to
agentsgen understand(--compact-budget, optional--focus, optional--changed) with explicit policy modes:full,focus,changed,focus+changed - owns the first central registry baseline for registered repos
- compares expected registry-derived
set.ymlagainst local repo workflows in read-only mode
SET now owns the first real repo-config contract and central registry baseline.
- Contract docs:
docs/repo-config.md - Canonical schema:
schema/repo-config.v1.json - Example config:
examples/repo-config.example.json - Central registry (first home):
registry/repos/*.json - Validate locally:
python3 scripts/validate_registry.py agentsgen.repomap_policylets each repo set compact budget, ranked-file limits, and optional focused/changed slice defaults without changing the Action contractagentsgen.proof_looplets a repo opt into contract/evidence/verdict artifacts for larger tasks, including evidence status, blocker counts, review readiness, and optional expected-artifact blockers- Derived policy modes in SET vocabulary:
full(Full Repo Slice),focus(Focused Code Slice),changed(Changed Files Slice),focus+changed(Hybrid Slice)
Planning-only helper for future PR-based config apply:
- Docs:
docs/config-apply-planning.md - Command:
python3 scripts/plan_config_apply.py markoblogo/lab.abvx - JSON mode:
python3 scripts/plan_config_apply.py markoblogo/lab.abvx --format json - Review bundle export:
python3 scripts/plan_config_apply.py markoblogo/lab.abvx --export-dir /tmp/set-plan - gh-ready payload: includes
base,head,title, andbody_filefields for a latergh pr createstep - apply simulation: previews branch name, target file write, commit message, and manual apply commands
- batch mode: accepts multiple repos or
--allfor a planning-only multi-repo summary with status/priority hints - operator hints: planner payload now includes
apply_readiness,operator_queue,blocked_by, structured capabilitywiring_gaps,next_action_label,recommended_operator_step, andnext_shell_commandfor UI/operator flows - workflow loop closure:
python3 scripts/plan_config_apply.py markoblogo/lab.abvx --repo-root /absolute/path/to/repocompares expectedset.ymlfrom registry against the real repo workflow and reportsmatches,drift, ormissing - repomap policy now flows through planning output so downstream control-plane views can show compact status and top ranked files per repo
- Keep the GitHub Action thin
- Keep registry and planner contracts explicit
- Use reviewable planning and drift checks before any future apply automation
docs/llmo-capability-map.mddocs/v0.1-scope.md