How it works
From markdown promises to enforced contracts — in three chapters.
Chapter 1
What Red Tape is
A briefing service
Every agent onboards by pulling a living briefing — mission, charter, guardrails, teammates, recent history — from one endpoint. It is never stale, because it's assembled from the ledger at request time.
An enforcement layer
Handoffs and guardrails aren't suggestions in a markdown file. The API rejects lazy session closes and denies destructive actions on protected paths — with your reason attached.
A system of record
Sessions, ledger entries, policy decisions, briefing reads — everything lands in an immutable audit trail. Your agent workforce becomes as accountable as your human one.
Chapter 2
A working day on the ledger
Six moments — from the incident that started it to the audit trail that ends it. Sit back; it plays itself.
01 / 06
Born from a real incident
An agent decided an untracked folder was disposable and cleaned it. The postmortem became a product: guardrails that agents must consult before acting.
audit_events
policy.deny · agent tried: delete C:\DEV\_AI
reason: "Untracked does not mean disposable."
incident prevented ✓
Map your company
Create a workspace per team or silo. Give each a mission, a charter, and an accountable owner. Register your agents and protect the paths that must never be touched.
Point your agents at the API
Agents open a session, pull their briefing into context, and check policy before anything destructive. A few curl calls — works with Claude Code, Codex, OpenClaw, or anything custom.
Handoffs enforce themselves
No session closes without a real handoff summary, files touched, blockers, and a changelog entry. The next agent onboards with full situational awareness.
Chapter 3
What it's worth to your workflow
Minutes, not mornings
Agent onboarding drops from hand-curating context files to one API call. The briefing is always current, so you stop paying the re-discovery tax on every session.
Zero cleanup incidents
The expensive failure mode of agent fleets — a destructive action on a folder that mattered — is blocked at the API before it happens, not discovered in a postmortem after.
Audit-ready by default
Compliance, client reporting, or just knowing what your fleet did: the ledger is the receipt. No reconstruction, no guesswork.
Before — governed by markdown
- ✕Context lives in AGENTS.md files that rot the moment they're written
- ✕Handoffs are a convention agents skip under pressure
- ✕Guardrails are vibes — one eager git clean and a folder is gone
- ✕“What did the agents do last week?” means grepping chat logs
- ✕Every new agent session starts by re-discovering the company
After — behind the tape
- ✓Briefings assembled live from the ledger on every onboard
- ✓Session close is rejected (HTTP 422) until the handoff is real
- ✓Destructive actions checked first; protected paths deny with your reason (HTTP 403)
- ✓One audit feed: every actor, action, and timestamp
- ✓Every session starts with the company map and what happened before it arrived