Agent · Dev 💻

E11 Dev (skill: e11-agent-dev) — the implementation engineer. Executes stories, drives the quick flow for small work, launches unattended runs, and reviews code. Alloy discipline is baked into its persistent facts: pre-load before editing @engineering11/* code, cite what you applied, be honest about what you skipped.

What it produces

• Code + tests in the target repo, scoped to the story's tasks or the quick-flow spec.
• An updated story file — tasks marked [x] only when tests pass, Dev Agent Record with the "Alloy pre-load complete" evidence, File List, Change Log.
• Spec files (quick flow) — spec-<slug>.md under {implementation_artifacts} with status frontmatter.
• Review findings written into the spec/story when [CR] runs.

When to use it

• A story is ready-for-dev on the sprint board — [DS].
• Small build/fix/refactor that doesn't deserve story ceremony — [DF] (quick flow, oneshot for tiny changes).
• You want the story implemented while you're at lunch — [DA] (unattended; halts blocked instead of asking).
• Code needs a fresh-context review — [CR].

How to invoke

Invoke e11-agent-dev; it greets and presents its menu:

Code	Dispatches to	Purpose
[DS]	e11-dev-story	Execute a prepared story, tests-first, in order
[DF]	e11-dev-flow	Quick flow — plan/implement/review, or oneshot
[DA]	e11-dev-auto	One fully unattended iteration; HALTs done/blocked
[CR]	e11-code-review	Layered adversarial review + triage
[QK]	e11-knowledge	Pull an alloy doc directly

The alloy pre-load (capability-driven)

Before touching code, e11-dev-story step 2b resolves which platform docs matter — and not just from imports:

1. Derives the story's task flavor (backend / frontend / shared) and its capabilities via skill:e11-knowledge pattern:backend-capability-map (+ pattern:forbidden-backend-patterns).
2. Maps capabilities to SDKs, then unions that with every @engineering11/* found in the story's SDK stamp (**E11 SDKs touched:**, written by e11-create-epics-and-stories), Dev Notes, tasks, and the architecture composition table.
3. Loads each SDK's doc (sdk:<name>:backend-using / :frontend-using), falls back to package source at node_modules/@engineering11/<name>/src/, and halts on a gap: [R]ead source / [S]kip convention check / [A]scalate.
4. Writes an "Alloy pre-load complete" record (capabilities, SDK lists, counts, alloy:// cites) into the Dev Agent Record — the exact evidence e11-code-review's alloy auditor later verifies.

Example session

You: e11-agent-dev

Dev: 💻 E11 Dev. Board says story 2-3-tenant-switch is ready-for-dev.

You: DS

Dev: Loaded 2-3-tenant-switch.md. Stamp: @engineering11/auth, access, user.
     Capability pass: pattern:backend-capability-map → +datastores (counters).
     Alloy pre-load complete: 4 SDKs (4 alloy-covered, 0 gaps). Cites recorded.
     Task 1/4: TenantSwitch handler ... tests 24/24 ✓ marked [x]
     ...
     Story → review. Run CR in a fresh context.

Gotchas

• The story file is authoritative. Dev implements tasks in order and refuses to mark [x] without passing tests. Fix the story, not the agent.
• Skips are visible. Choosing [S]kip at a gap writes a skip marker the alloy auditor will surface (AUDIT-HIDDEN-SKIP if it's concealed). That's the point.
• [CR] in a fresh context. The review orchestrator assigns severity only after reading surrounding code — give it a clean session, ideally a stronger model.
• [DA] never asks. If intent is unclear it HALTs blocked with the reason written into the spec file. Sharpen the intent and re-run.

Customizing

Persona/menu/facts: .foundry/custom/e11-agent-dev.toml. Workflow behavior belongs to the workflow skills — e.g. review layers in .foundry/custom/e11-code-review.toml, the e11_platform_preload gate in .foundry/custom/e11-dev-story.toml. See Customization.

Source

src/foundry/workflows/4-implementation/e11-agent-dev/SKILL.md

See also: Phase 4 — Implementation · Quick flow · Unattended runs · Ship a story end-to-end.