"`. Architecture docs and security policies are common hits at this stage. If both providers are off (default), skip silently. ## Exit criteria - File written at `fastpace/docs/erd/<slug>.md` with `status: draft`. - Every non-trivial design choice has a citation. - Security section non-empty. - Risks & Open questions non-empty (or explicitly "none — and here's why"). - Threat model generated (`fastpace/threats/<slug>.md`) and every open STRIDE gap either mitigated in the ERD or accepted under Risks. (Skip only if the command is unavailable — note that in the ERD.) - Compliance check run if the PRD declares a regulatory framework, and every open control addressed or noted out-of-scope. - Output: file path + acceptance-criteria count + threat-model open-gap count + suggested next command.

` ``` ## Threat model & compliance check Once the ERD file exists, run these two commands (idempotent; safe to re-run with `--force` on the threat model). Both read the security section you just wrote and emit signed audit events. ```bash fastpace threat-model --force fastpace compliance check-erd ``` - **`threat-model`** seeds a STRIDE + LINDDUN model at `fastpace/threats/.md` from the signals in your Security section, and flags any STRIDE category that *applies but has no mitigation* as an open gap. For each open gap it reports, either (a) add the missing mitigation to the ERD's Security section, or (b) record it as an explicit accepted risk under **Risks & open questions** with a rationale. Do not leave a flagged gap silent. - **`compliance check-erd`** resolves the frameworks from the PRD's `**Regulatory:**` constraint (or pass `--framework `), pulls the mapped controls from `framework-mapping.md`, and writes `fastpace/compliance/erd-.md` listing which controls the ERD addresses vs. leaves open. For each **open** control, add a sentence to the Security section covering it, or note in **Risks** why it's out of scope for this feature. If the PRD declares no regulatory frameworks, `check-erd` will say so — that's fine; skip it. If either command is unavailable (license tier / not installed), skip silently and note in the ERD that the threat model / compliance check was not run. When you show the draft (step 6), summarize for the user: STRIDE categories that apply, any open threat gaps, and per-framework addressed/open control counts. ## Citation rule For any non-trivial design choice, cite the context file that drove it. Examples: - Choosing retry policy → cite `patterns.md § retry-with-backoff`. - Not logging PAN → cite `decisions.md § 2025-02-14`. - Putting code in `api/cart/` → cite `architecture.md § modules`. - Avoiding a repeated bug → cite `learnings.md § `. Citations in the ERD go in the body where the choice is made, not in a list at the end. Future readers see *why* next to *what*. ## Rules - **No ad-hoc library choices.** If the ERD introduces a new dependency (npm / pip / cargo), call it out explicitly: "New dependency: `@` — needed for ``. Alternatives considered: ``." The user must approve before you finalize. - **No code implementations.** The ERD is a design doc. Short pseudocode / signatures are fine; full function bodies are not. - **Security is not optional.** An empty security section fails the ERD. - **Ground every boundary call.** If you say "this goes in module X", cite architecture.md. - **Don't weaken acceptance criteria.** If the PRD says "reject cards with expired dates", the ERD can't silently drop that. - **Mark `TBD` instead of guessing.** Performance numbers, canary percentages, migration timing — if the user hasn't said, write TBD + add to Open Questions. ## Integration - After the ERD is written, tell the user: "Next: `/fp-erd-to-execplan ` once this is approved, to break it into waves/epics/stories." - The threat model at `fastpace/threats/.md` becomes an input to the risk register — mention it's available. - If you had to flag a decision conflict, suggest `/fp-write-adr` to formally supersede it. - If the ERD surfaced a brand-new pattern worth reusing, suggest `/fp-remember` for `patterns.md`. ## Auto-discovery (optional, off by default) If `references.resolvers.confluence.auto_discover.enabled` or `.google-docs.auto_discover.enabled` is true in `fastpace.config.yaml`, run after step 5: ```bash fastpace refs suggest --limit 5 ``` The query uses the ERD body you just wrote (architecture-heavy text — usually surfaces different docs than the PRD did, since ERDs include module names, library choices, security concerns). Show candidates as a numbered list and ask: > *"Add any of these to the ERD's references? Reply with numbers (e.g. `1, 3`) or `none`."* For each chosen item, run `fastpace refs add --url --title ""`. Architecture docs and security policies are common hits at this stage. If both providers are off (default), skip silently. ## Exit criteria - File written at `fastpace/docs/erd/<slug>.md` with `status: draft`. - Every non-trivial design choice has a citation. - Security section non-empty. - Risks & Open questions non-empty (or explicitly "none — and here's why"). - Threat model generated (`fastpace/threats/<slug>.md`) and every open STRIDE gap either mitigated in the ERD or accepted under Risks. (Skip only if the command is unavailable — note that in the ERD.) - Compliance check run if the PRD declares a regulatory framework, and every open control addressed or noted out-of-scope. - Output: file path + acceptance-criteria count + threat-model open-gap count + suggested next command.