Agent · Tech Writer 📚

E11 Tech Writer (skill: e11-agent-tech-writer) — the documentation specialist. Fluent in CommonMark, DITA, OpenAPI, and Mermaid. Optimized for "I need a document, not a by-product" requests, with a persistent standards memory that survives sessions.

What it produces

• Project docs, architecture writeups, API references, explainers.
• Mermaid diagrams — sequence, flowchart, ER, state.
• Validation reports — does this doc meet our standards?
• Doc indexes — via e11-index-docs.

When to use it

• A document is the deliverable, not the side-effect of another workflow.
• An existing doc needs review for clarity, structure, or convention adherence.
• You need brownfield project documentation ([DP]) with a writer's editorial hand.

How to invoke

Invoke e11-agent-tech-writer. Its menu mixes skill dispatch and prompt-driven commands (the four writing commands are prompt files bundled with the agent):

Code	Command	Kind	Purpose
[DP]	Document Project	skill → e11-document-project	Brownfield analysis → project docs
[WD]	Write Document	prompt	Produce a doc per the standards sidecar
[MG]	Mermaid Generate	prompt	Diagram from a description
[VD]	Validate Document	prompt	Review a doc against standards
[EC]	Explain Concept	prompt	Clear technical explanation with examples
[ID]	Index Docs	skill → e11-index-docs	Build an index.md for a docs folder
[QK]	Query Knowledge	skill → e11-knowledge	Pull an alloy doc directly

The standards sidecar

The Tech Writer loads .foundry/_memory/tech-writer-sidecar/documentation-standards.md as a persistent fact on every activation. When you correct a convention once ("we always spell it datastore"), have it update the standards file — every future session inherits the rule. This is the one agent with durable, file-backed memory by default.

Example session

You: e11-agent-tech-writer

TW: 📚 E11 Tech Writer. Standards sidecar loaded (14 conventions).

You: WD — a runbook for rotating the service-to-service signing keys.

TW: Scope check: audience ops engineers, format runbook (numbered,
    verification step after each action). Sources: I'll pull
    alloy://shared/infrastructure/cloud-run-conventions.md for deploy specifics.
    Draft at docs/runbooks/rotate-s2s-keys.md — validating against
    the sidecar standards... 2 fixes applied. Done.

Gotchas

• Prompt commands are agent-local. [WD]/[MG]/[VD]/[EC] live inside the agent-skill directory — they're not standalone skills you can invoke without the agent.
• Editorial reviews are separate skills. For a clinical copy-edit or a structural pass, use e11-editorial-review-prose / e11-editorial-review-structure — the writer can invoke them, and the PRD/brief finalize gates use them as doc_standards reviewers.
• Keep the sidecar in the repo. It lives under .foundry/_memory/ in your project; losing it means re-teaching conventions.

Customizing

Persona, menu, and facts: .foundry/custom/e11-agent-tech-writer.toml. Add org writing rules as persistent_facts entries, or point a menu item at your own prompt file. See Customization.

Source

src/foundry/workflows/1-analysis/e11-agent-tech-writer/SKILL.md

See also: Phase 1 — Analysis · Agent quick-reference.