Capability Deltas
Before, enabling change, after, human consequence, verification, and buildability in one inspectable record.
3 observed deltas
/deltas.jsonOpen JSON indexDateCapability deltaTransitionHuman consequenceVerificationBuildabilityRoutes
01Explicit fields, precise errors, request metadata, and raw facts turn an agent retry from guessing into near-deterministic repair.
BeforeAgent calls a human-friendly SDK and receives an opaque failure.
Enabling changeAPI returns command, exit code, stderr, allowed values, request id, timeout, and raw metadata.
AfterAgent can correct the next tool call from explicit machine-readable facts.
The human spends less time reading logs for the agent and explaining what exactly broke.
source-inspectedB2engineer-repeatable
02A security agent can start from attacker-first exploit-path search and prepare a production-code fix instead of following an abstract checklist.
BeforeSecurity review identifies a vulnerability and hands remediation back to humans.
Enabling changeVulnHunter-style agent reasons from attacker behavior to vulnerable code and repair candidates.
AfterThe workflow can move from exploit-oriented analysis to proposed code change faster.
Security teams get a more actionable review artifact, but still need human approval before code changes.
source-inspectedB1bespoke demo
03A separate context layer lets multiple AI clients read shared project context without a human manually copying memory.
BeforeEach AI client learns the project separately and loses context when the user switches tools.
Enabling changeA shared MCP-compatible context layer exposes structured memory outside one client.
AfterDifferent agents can reuse the same governed project context.
The user spends less time carrying decisions, architecture notes, and preferences between Claude, Codex, Cursor, and other clients.
source-inspectedB1bespoke demo
No deltas match these filters.