ContextECF Suite
One evidence trail for
AI-assisted engineering.
CodeLedger, FabricLedgerSDK, and CFA are three separately versioned products that solve three different problems. Together, they let you trace a single piece of AI-assisted work all the way from the task to the files it touched, to what the model actually saw, to whether it was allowed to act — with a receipt at every step.
Three layers. One shared trail.
Each component keeps its own release cycle and runtime boundary — this isn't a monolith.
CodeLedger
Picks the right files.
Scans your repo and hands your AI agent a precise, ranked bundle of the files that actually matter for the task — before it goes looking on its own.
FabricLedgerSDK
Cleans up the prompt.
Prunes stale docs and copied terminal noise out of the prompt, and strips or flags unsafe instructions — like "ignore the tests" hidden in a file — before the agent ever sees them.
CFA
Enforces the policy.
The enterprise gateway: approvals, manifest enforcement, and a permanent record of anything the agent tried to do that policy denied.
The transaction spine
Every layer attaches evidence to the same ID.
CodeLedger opens the engineering task transaction. FabricLedgerSDK attaches its prompt/runtime decisions to that same transaction. CFA attaches its gateway decisions to it too. The result is one traceable line, not three disconnected logs.
How teams actually adopt it
Nobody needs to understand the whole Suite on day one. It's a staged path, not a bundle purchase.
Start with CodeLedger
Give your coding agents fresh, relevant task context and a local receipt for what it saw. This alone is most of the value for most teams.
Add FabricLedgerSDK
When your team wants the prompt itself cleaned up — noise pruned, bad instructions filtered — before it reaches the model or a tool executes.
Add CFA
When the organization needs managed policy, human approvals on risky actions, and an audit trail that satisfies a security or compliance review.
Ready to level up?
Go from "installed CodeLedger" to the full Suite.
One command tells you exactly what's missing and what to run next — codeledger contextecf status. From there, contextecf install fabric and contextecf install cfa hand you the real install command for FabricLedgerSDK's prompt pruning and CFA's enterprise gateway — no guesswork, and nothing runs on your machine until you say so.
What this looks like in practice
Three illustrative scenarios — not real customer data.
Login rate-limit fix
CodeLedger
Selects the auth route, middleware, and tests — skips the unrelated 40% of the repo.
FabricLedgerSDK
Keeps stale docs and copied terminal noise out of the prompt.
CFA
Not required — this change doesn't touch policy-gated auth config.
Outcome: Focused context, reviewable receipt, no wasted agent time.
Prompt injection hidden in repo docs
CodeLedger
Still identifies the correct docs and code paths for the task.
FabricLedgerSDK
Flags and strips the "ignore the tests" instruction before the agent reads it.
CFA
Logs a denied-capability record if the unsafe command is attempted anyway.
Outcome: The agent gets useful context without ever obeying the injected instruction.
Payment gateway change
CodeLedger
Opens the task transaction, records validation evidence.
FabricLedgerSDK
Records prompt/runtime evidence without exporting raw prompts.
CFA
Requires human approval before the deploy/write action is allowed to run.
Outcome: A reviewer can trace task → context → prompt → gateway decision, end to end.
Local-first by default
Source code doesn't leave the machine by default, and raw prompts aren't exported by status or receipt commands. Each component keeps its own local state (.codeledger/, .fabricledger/, .cfa/).
Get started
Not sure where to start?
Tell us what you're working with and we'll help you figure out the right layer to start with — CodeLedger alone is often enough on day one.