refactor: overhaul skill discovery and authorized workflows

AGENTS.md

@@ -0,0 +1,29 @@
+# Working on Skills
+
+`parsa/` contains the active Claude, Codex, business, and SEO workflows.
+`tyler/` is a frozen ancestor; Orchestra's canonical source is dcouple/orchestra.
+Edit active sources here, preserving harness-specific tools and invocation
+metadata. Shared variants that are identical should remain identical.
+
+## Authoring and completion
+
+Keep descriptions concise and discriminating. Put conditional detail in linked
+references with a read condition. Preserve output formats, caller names,
+explicit authorization boundaries, and the evidence that determines readiness.
+Load only the affected skills, callers, and relevant supporting material.
+
+For implementation requests, continue through scoped changes, checks, fixes,
+and the requested PR. Local branches, worktrees, and disposable verification
+fixtures are part of that work. Existing authorization persists across routine
+stage boundaries; planning-only requests and explicit review pauses still stop
+before implementation. Merge, release, deployment, production/destructive
+changes, and external posting require their own matching authorization.
+
+Validate changed metadata, local reference links, and install/sync behavior.
+Run scripts only against disposable fixtures for this audit, not the user's
+installed skills or consumer repositories. Reuse passing results while inputs
+remain unchanged. Read back pushes and PR state before reporting success.
+
+Keep `./tmp/` scratch and unrelated changes out of commits. Discover repository
+commands and target environments instead of hardcoding a consumer's stack.
+Do not use em dashes in new instructions or PR prose.

CLAUDE.md

@@ -1,22 +1,5 @@
-## Rules
+See @AGENTS.md for scope, authoring, validation, and authorization boundaries.
 
-- **You must verify after every mutation.** After a merge, push, PR
-  creation, file move, or any operation that changes state: read the
-  actual result back. Never trust command output alone - check that the
-  change landed where you expected it to.
-
-- **You must not solve discoverable problems.** If the agent can query it
-  at runtime (via MCP, the environment, or the repo), don't hardcode it.
-  Describe roles, boundaries, and rules - not configuration.
-
-- **Less is more.** Every line should earn its place. If removing a line
-  wouldn't change behavior, remove it. Overspecification leads to
-  configuration ceremony, leads to brittleness. When things are brittle,
-  they break.
-
-- **No feedback loops.** A step that mutates code after review invalidates
-  downstream work. The pipeline is linear. Refactoring and cold-read are
-  manual skills, never inline.
-
-- **You must never use em dashes.** Use commas, periods, colons, or
-  parentheses instead.
+Fixes after review invalidate affected downstream evidence. Final QA belongs
+after the last reviewed code change. Refactoring and cold-read are requested
+workflows, not automatic implementation gates.

docs/README.md

@@ -0,0 +1,40 @@
+# Workflow visuals
+
+The current Parsa workflows live in `parsa/`. Current Orchestra is maintained
+in [dcouple/orchestra](https://github.com/dcouple/orchestra). The two historical
+images below retain the old design for context and carry visible archive
+labels so they cannot be mistaken for installation or execution instructions.
+
+| Visual | Status and scope | Source | Render |
+| --- | --- | --- | --- |
+| Parsa workflow | Current software stages, continuation, model selection, and business handoffs | [Excalidraw](readme-workflow-map.excalidraw) | [PNG](readme-workflow-map.png) |
+| Skill legend | Current primary skills and supporting roles; installation can add a `p-` prefix to collisions | [Excalidraw](readme-skill-legend.excalidraw) | [PNG](readme-skill-legend.png) |
+| SEO workflow | Campaign strategy, focused page work, writing-skill selection, and publication boundaries | [Excalidraw](seo-workflow-map.excalidraw) | [PNG](seo-workflow-map.png) |
+| Software-factory story | Conceptual evolution; paired with Orchestra's copy in this overhaul | [Excalidraw](software-factory-story.excalidraw) | [PNG](software-factory-story.png) |
+| Tyler workflow | Historical snapshot of the frozen ancestor; old capture names and model routes are retained as history | [Excalidraw](tyler-workflow-map.excalidraw) | [PNG](tyler-workflow-map.png) |
+| Software Orchestra sketch | Historical early design, including proposed loops and templates | [Excalidraw](software-orchestra.excalidraw) | [PNG](software-orchestra.png) |
+
+The software-factory story describes a direction for configured intake. It
+does not imply that every pictured schedule or deployment is running. Use
+[Orchestra's workflow guide](https://github.com/dcouple/orchestra/blob/main/WORKFLOW.md)
+for its current executable contracts.
+
+## Regenerate
+
+From the repository root, install the renderer dependencies once:
+
+```bash
+uv sync --project parsa/.claude/skills/excalidraw-pr-diagrams/references
+uv run --project parsa/.claude/skills/excalidraw-pr-diagrams/references playwright install chromium
+```
+
+Render each changed source and inspect its PNG at README size. Fix clipping,
+overlap, or wrong connections, then commit the source and PNG together:
+
+```bash
+uv run --project parsa/.claude/skills/excalidraw-pr-diagrams/references python parsa/.claude/skills/excalidraw-pr-diagrams/references/render_excalidraw.py "$PWD/docs/readme-workflow-map.excalidraw"
+```
+
+Replace the input filename for the other maps. The `.codex/` variant carries
+the same rendering tools. Historical content stays historical; refresh its
+archive label or links if needed without presenting it as today's workflow.