docs: fix skill counts, crate list, memory status and telemetry wording

README.md

@@ -17,8 +17,9 @@ Three ideas shape it:
 
 - **You own the whole thing.** Your key, your model, your machine. Sessions,
   memory, and skills are plain files on your disk. There is no hosted control
-  plane, no vector database, and no telemetry. Point it at a local model and it
-  works with no network at all.
+  plane, no vector database, and no telemetry unless you point it at your own
+  OpenTelemetry endpoint. Point it at a local model and it works with no network
+  at all.
 - **A harness, not a prompt.** Tools, permissions, history, memory, and
   retrieval are shared infrastructure. Chat, review, and fix all inherit them,
   so a new capability is a small addition rather than another agent rebuilding
@@ -372,8 +373,9 @@ listed by title and read in full only when the agent needs them.
 Skills are folders with a `SKILL.md` telling the agent how to do something
 specific. Aster reads the titles and loads the body only when it is relevant.
 
-Five core skills ship built in and are always available: git and GitHub
-workflows, security review, security scanning, and skill authoring. Eleven more
+Six core skills ship built in and are always available: git and GitHub
+workflows, security review, security scanning, skill authoring, and artifact
+design. Twelve more
 guide the agent's own conduct (verifying before reporting done, build triage,
 shell batching, CLI craft, context economy, web research, taking corrections,
 and so on); those are internal, so they never appear in a skills list or as a
@@ -559,6 +561,8 @@ crates/
   aster-eval/        the evaluation harness
   aster-telemetry/   optional OpenTelemetry export
   aster-shortcuts/   Apple Shortcuts tools, listed and run without opening the app
+  aster-voice/       voice dictation: records the microphone and transcribes it
+  aster-jev/         experimental Jev advice for the agent loop, behind the `jev` feature
   aster-models/      shared domain types
 desktop/             the desktop app (Tauri)
 editors/vscode/      the VS Code extension, and the panel `aster serve` hands a browser

aster.yaml.example

@@ -134,9 +134,9 @@ agents:
 ui:
   # Print the session header (model, provider, skills) when chat starts.
   welcome: true
-  # Chat color theme: dark, light, midnight, or forest. `/theme` in a chat
-  # picks one and saves.
-  theme: dark
+  # Chat color theme: default, light, midnight, forest, or any theme in
+  # THEMES.md. `/theme` in a chat picks one and saves.
+  theme: default
 
 # MCP servers that give the agent extra tools. `aster mcp list` shows what they
 # expose; `aster mcp enable|disable <name>` flips them without editing by hand.

docs/MEMORY.md

@@ -262,8 +262,6 @@ recorded as an `eviction` event, so the transcript shows what was dropped.
 
 ## What is deferred
 
-- **Auto end-of-session distillation** of memory: save what is relevant to the
-  work and the user, not every fact stated (memory is tool-driven and manual today).
 - **A derived SQLite/FTS5 recall index** over transcripts and memory, added only
   when a linear scan across files gets slow. It would be rebuildable from the
   files, never authoritative.
@@ -297,12 +295,14 @@ Implemented incrementally; each stage lands independently:
   `MAX_INDEX_ENTRIES` (60), ordered most-recent-first, with a truncation notice.
   Model-visible memory cannot grow without bound.
 
-### Stage 2 (next): consolidation
+### Stage 2: consolidation
 
-A model-backed consolidation pass, run on turn/session boundaries, distills
-durable semantic memory from the episodic transcript: propose new blocks,
-merge duplicates, drop contradictions, archive stale facts. Bounded and
-budgeted per AGENTS.md; never dumps the full transcript into context.
+When a session of at least six turns finishes, a model-backed pass
+(`aster-memory`) reads a bounded digest of the transcript and proposes new
+blocks, merges of duplicates, archives of facts the session contradicted, and
+lessons from corrections. Every change goes through the journal, and a marker
+keeps a session from being consolidated twice. It never puts the full
+transcript into context.
 
 ### Stage 3 (next): retrieval