Essential session storage is required to sign in. Optional analytics remain off unless you allow them.
Deliver an end-to-end case study and working critical path for adopting coding agents across a team. Requirements: (1) map developer workflows and identify where agents create measurable value versus unacceptable risk; (2) define a permission model for repository reads, edits, commands, external tools, CI, and production boundaries; (3) create project instruction hierarchy and change-control process; (4) evaluate at least one external integration or MCP server for authentication, data scope, write authority, auditability, idempotency, approval, rollback, and deferral conditions; (5) implement a bounded non-interactive CI analysis artifact with structured output, redacted untrusted input, timeout, budget, error behavior, and no autonomous merge or deploy authority; (6) demonstrate subagent delegation for independent read-only investigations with one integration owner; (7) produce security threat model, incident runbook, and rollout plan; (8) define adoption, quality, cost, and failure metrics plus a post-pilot decision framework.
The workflow map, task framing, instruction hierarchy, permission model, review path, and governance owners form a coherent and usable team operating model.
Integrations, untrusted input, secrets, CI authority, typed outputs, external writes, audits, incident response, rollback, and escalation are designed with controls proportionate to impact.
The project provides realistic pilot boundaries, verification evidence, adoption and cost metrics, failure signals, support process, and explicit criteria for expanding, narrowing, or stopping the program.
Use this README structure to make the work reviewable.
DevPath fetches a read-only, size-capped snapshot. Submitted code is never executed.
# A team-safe agentic development platform ## Goal Deliver an end-to-end case study and working critical path for adopting coding agents across a team. Requirements: (1) map developer workflows and identify where agents create measurable value versus unacceptable risk; (2) define a permission model for repository reads, edits, commands, external tools, CI, and production boundaries; (3) create project instruction hierarchy and change-control process; (4) evaluate at least one external integration or MCP server for authentication, data scope, write authority, auditability, idempotency, approval, rollback, and deferral conditions; (5) implement a bounded non-interactive CI analysis artifact with structured output, redacted untrusted input, timeout, budget, error behavior, and no autonomous merge or deploy authority; (6) demonstrate subagent delegation for independent read-only investigations with one integration owner; (7) produce security threat model, incident runbook, and rollout plan; (8) define adoption, quality, cost, and failure metrics plus a post-pilot decision framework. ## Acceptance criteria - [ ] End-to-end operating model: The workflow map, task framing, instruction hierarchy, permission model, review path, and governance owners form a coherent and usable team operating model. - [ ] Security and automation discipline: Integrations, untrusted input, secrets, CI authority, typed outputs, external writes, audits, incident response, rollback, and escalation are designed with controls proportionate to impact. - [ ] Evidence-driven rollout: The project provides realistic pilot boundaries, verification evidence, adoption and cost metrics, failure signals, support process, and explicit criteria for expanding, narrowing, or stopping the program. ## Verification - [ ] Document installation and run commands. - [ ] Record automated checks and their exact commands. - [ ] Add representative output, screenshots, or a short demo where useful. - [ ] Confirm failure paths and known constraints. ## Decision log ### Decision title - Context: - Choice: - Alternatives considered: - Tradeoffs: ## Evidence - Link each rubric criterion to the file, test, or artifact that demonstrates it. ## Known limitations - List what is intentionally out of scope and what should be improved next.