System repo for running a compounding agentic R&D loop:

1. Capture work as evidence.
2. Distill evidence into reusable context.
3. Evaluate with explicit contracts.
4. Promote or reject policies.
5. Repeat with receipts.

• docs/ architecture, roadmap, operating model
• loops/ run definitions and loop specs
• benchmarks/ eval suites and fixtures
• policies/ governance rules and gates
• receipts/ run artifacts and audit records
• dashboards/ visualization surfaces
• state/ normalized state snapshots
• ops/ war-room procedures
• scripts/ automation utilities

Ship one end-to-end loop that can:

• ingest context from prior sessions,
• run a task with policy gates,
• produce receipts,
• score outcomes against benchmarks,
• update a state snapshot.

Default run (uses Codex history status report as context):

python3 /Users/jobs/Desktop/agentic-os/scripts/run_core_loop.py

Custom task:

python3 /Users/jobs/Desktop/agentic-os/scripts/run_core_loop.py \
  --task-file /Users/jobs/Desktop/agentic-os/tasks/sample_task.json

Outputs:

• Receipts: receipts/runs/<run_id>/
• State snapshot: state/latest.snapshot.json
• Workspace scan evidence: receipts/runs/<run_id>/inputs.json under workspace_scan

Run system task pack:

bash /Users/jobs/Desktop/agentic-os/scripts/run_task_pack.sh

System tasks included:

• tasks/nix_codecli_operational_hardening.json
• tasks/meta3_testbed_capability_ledger.json
• tasks/graphlogue_revival_slice.json

Scan enforcement:

• Each task can define workspace_roots, scan_markers, required_markers_any, and min_existing_roots.
• Policy gates fail when scan requirements are not met.

Run hologram dogfood lane:

bash /Users/jobs/Desktop/agentic-os/scripts/run_hologram_dogfood.sh

This lane enforces evidence checks for:

• strategy declaration (HOLOGRAPHIC_STRATEGY.md)
• hologram sensor wiring (src/api.rs)
• negentropy hologram mode (src/bin/negentropy.rs)

Generate a live chat runtime proof receipt:

bash /Users/jobs/Desktop/agentic-os/scripts/prove_live_chat_receipt.sh

This writes:

• smoke log: receipts/intent_reports/live_chat_smoke_<timestamp>.log
• proof metadata: receipts/intent_reports/live_chat_smoke_<timestamp>.json