ONE LAYER · TWELVE CAPABILITIES
Start with eight free jobs during delivery. Continue with four planned operated services after go-live. Every workflow shows the input, action, artifact, work avoided, and what it cannot know.
ENTRY POINT · NOT SOLUTION 0
The local control surface through which an FDE would choose, run, and inspect the free layer inside a repository. Toggle to the CLI path to see the same repository and qualified command evidence.
adopt ui is not available in the public CLI. This simulation executes nothing.
You no longer redraw the delivered architecture for the handover deck.
Action status: Planned local interface — illustrative.
Limit: The Blueprint cannot infer undocumented intent, unseen runtime behavior, or dynamic references beyond the declared observation boundary.
You no longer reconstruct the decision from memory or old chat threads months later.
Action status: Planned workflow — illustrative.
Limit: When recoverable evidence does not explain the rationale, Decision Memory marks it unknown and asks a human to confirm it.
You stop repairing guidance merely because a file, screen, or symbol moved.
Action status: Planned local interface — illustrative.
Limit: Ambiguous, generated, and runtime-only references can require review; matching an identity does not prove semantic equivalence.
You leave a repeatable reference for detecting behavior change instead of relying on memory.
Action status: Planned workflow — illustrative.
Limit: A baseline is not production monitoring or proof of correctness; it covers only approved probes and cannot trigger real side effects.
Routine project questions start with inspectable evidence instead of returning immediately to you.
Action status: Planned local assistant — illustrative.
Limit: The assistant is not a generic chatbot or support desk; it refuses unsupported, stale, or out-of-bound questions.
You review a generated handover instead of authoring a deck and runbook from scratch.
Action status: Planned workflow — illustrative.
Limit: The generated package is a reviewable draft; it cannot fill unknowns or transfer operational accountability by itself.
You spend scarce subject-matter-expert time correcting missing areas instead of reviewing everything.
Action status: Planned local interface — illustrative.
Limit: Completeness is relative to the observed inventory and declared boundary, never a claim that the whole real-world system is documented.
You can state the evidence boundary clearly instead of letting confidence imply coverage.
Action status: Planned local interface — illustrative.
Limit: The statement reflects declared access and current observations; it is not a certification, compliance attestation, or proof about unseen systems.
adopt map . --firm acme --engagement support-agent --system acme-support-agent --archetype ai --jsonLimit: The Blueprint cannot infer undocumented intent, unseen runtime behavior, or dynamic references beyond the declared observation boundary.
# No public CLI workflow for the planned Decision Memory experienceLimit: When recoverable evidence does not explain the rationale, Decision Memory marks it unknown and asks a human to confirm it.
adopt identity build --scope acme/support-agent/acme-support-agent --kind prompt --namespace prompts --key refund-triageLimit: Ambiguous, generated, and runtime-only references can require review; matching an identity does not prove semantic equivalence.
adopt probe manifest validate evals/probes/refund-routing.yamlLimit: A baseline is not production monitoring or proof of correctness; it covers only approved probes and cannot trigger real side effects.
adopt agent check --adapter fake_recorded --jsonLimit: The assistant is not a generic chatbot or support desk; it refuses unsupported, stale, or out-of-bound questions.
adopt export ./onboard-exportLimit: The generated package is a reviewable draft; it cannot fill unknowns or transfer operational accountability by itself.
adopt coverage recompute --system acme-support-agent --jsonLimit: Completeness is relative to the observed inventory and declared boundary, never a claim that the whole real-world system is documented.
adopt boundary --answers adopt/boundary-answers.json --archetype ai --write-statement ./onboard-output/coverage-boundary.mdLimit: The statement reflects declared access and current observations; it is not a certification, compliance attestation, or proof about unseen systems.
Inspect this repository using the public adopt CLI. Start with `adopt detect ./acme-support-agent --json`. Report what the command verifies, what remains unknown, and every local artifact created. Do not send repository content over the network.
adopt ui is not a public command. The browser Workbench shown here is a planned experience. Verified commands are labelled separately.CAPABILITY CATALOG
Free capabilities prioritize the FDE: inspect the repository, create a concrete artifact, and reduce handover work while evidence is still close.
Nobody has a reliable structural map of what the engagement actually delivered.
You no longer redraw the delivered architecture for the handover deck.
Watch it workImportant rationale is scattered across pull requests, tickets, commits, and the memories of the delivery team.
You no longer reconstruct the decision from memory or old chat threads months later.
Watch it workGuidance tied to a screen position or file path breaks as soon as the implementation moves.
You stop repairing guidance merely because a file, screen, or symbol moved.
Watch it workThe delivered AI workflow has no safe, repeatable record of what its important behavior looked like at go-live.
You leave a repeatable reference for detecting behavior change instead of relying on memory.
Watch it workProject questions keep returning to the delivery team because the repository cannot explain itself with evidence.
Routine project questions start with inspectable evidence instead of returning immediately to you.
Watch it workThe map, rationale, tests, owners, and gaps are scattered when the FDE needs to produce a client handover.
You review a generated handover instead of authoring a deck and runbook from scratch.
Watch it workDocumentation exists, but its presence does not reveal which delivered surfaces remain uncovered or stale.
You spend scarce subject-matter-expert time correcting missing areas instead of reviewing everything.
Watch it workAn assistant can appear authoritative even when parts of the delivered system sit outside its evidence boundary.
You can state the evidence boundary clearly instead of letting confidence imply coverage.
Watch it workPlanned human and service workflows continue the same system memory. They are not separate upsells or substitutes for support, observability, ITSM, or maintenance.
Post-go-live questions repeatedly return to the original implementation team because no current, owned answer path exists.
A difficult question is answered once and the confirmed result becomes reusable instead of repeatedly reaching you.
Watch it workRepository, prompt, model, and configuration changes can make client guidance unsafe without anybody noticing.
You confirm or correct a prepared update instead of auditing and rewriting the knowledge set.
Watch it workGiving the receiving team a folder does not prove that ownership transferred or that realistic operating tasks can be completed.
You get a defined, evidence-backed exit path instead of an open-ended knowledge-transfer tail.
Watch it workPractice leaders cannot see which completed engagements still generate stale knowledge, unanswered questions, and ownerless review work.
Leaders can see where completed engagements still consume delivery capacity and assign ownership before escalation.
Watch it work