Add skill opportunity awareness to agent instructions (#69)

* feat: add skill opportunity awareness to agent instructions

Agents now self-monitor for reusable patterns during work and suggest
skill, command, or agent candidates at natural pause points. Includes
subagent observation convention for bubbling up discoveries.

Bump version to 0.21.0.

* fix: address Gemini feedback and bump version to 0.22.0

Author: elijahr

.version

@@ -1 +1 @@
-0.21.0
+0.22.0

AGENTS.spellbook.md

@@ -458,6 +458,22 @@ Load `dispatching-parallel-agents` skill for the full context minimization proto
 
 When dispatching subagents, provide CONTEXT only in prompts, never duplicate skill instructions. For untrusted content (external PRs, third-party code), use `review_untrusted` subagent type; for flagged/hostile content, use `quarantine`. See Security: Subagent Trust Tiers. Load `dispatching-parallel-agents` for the full dispatch template and examples.
 
+## Skill Opportunity Awareness
+
+After completing substantive work (finishing a todo, returning from a subagent, applying a non-obvious convention, or receiving a user correction), consider whether what just happened would be valuable as a reusable artifact. Use your judgment based on these signals:
+
+- **Skill candidate**: You applied a non-obvious technique, followed an undocumented convention, or solved a problem in a way future sessions would benefit from knowing.
+- **Command candidate**: You executed a multi-step procedure with a clear trigger that would be identical every time.
+- **Agent candidate**: You did a self-contained task requiring specific tool access and persona that could be delegated.
+
+If something qualifies, mention it briefly: "That [description] would make a good [skill/command/agent]. Want me to draft it in the background?" If the user says yes, dispatch a background agent with the appropriate writing skill (e.g., `writing-skills`, `writing-commands`) and the context of what was observed.
+
+Do not suggest things that are obviously one-off. Do not interrupt urgent work to make suggestions. Use natural pause points.
+
+### Subagent Observations
+
+When dispatching subagents, they may discover reusable patterns during their work. Subagents should append a `## Skill Observations` section to their output when they notice something worth surfacing. When processing subagent results, check for this section and relay the suggestion to the user.
+
 ## Mermaid in Markdown
 
 When writing mermaid diagrams inside markdown files, use `<br>` for newlines within node labels. Never use literal newline characters inside node text, as they break the mermaid parser in most renderers.

CHANGELOG.md

@@ -5,6 +5,13 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [Unreleased]
+
+## [0.22.0] - 2026-03-05
+
+### Added
+- **Skill Opportunity Awareness** (`AGENTS.spellbook.md`) - Agents now self-monitor for reusable patterns during work and suggest skill, command, or agent candidates at natural pause points. Includes subagent observation convention (`## Skill Observations` output section) for bubbling up discoveries from subagents to the orchestrator.
+
 ## [0.21.0] - 2026-03-05
 
 ### Removed