codecall 1.0.0

package/skill/references/AGENTS.codecall.md

@@ -0,0 +1,7 @@
+# Codecall automatic recommendation
+
+Before the normal final response for a completed implementation, read the
+installed codecall skill's `references/trigger-policy.md` and evaluate the
+change. When its outcome is `recommend`, present Start Learning / Skip and wait.
+Do not teach until Start Learning. For `optional` or `skip`, finish normally;
+`$codecall` always remains available.

package/skill/references/CLAUDE.codecall.md

@@ -0,0 +1,7 @@
+# Codecall automatic recommendation
+
+Before the normal final response for a completed implementation, read the
+installed codecall skill's `references/trigger-policy.md` and evaluate the
+change. When its outcome is `recommend`, present Start Learning / Skip and wait.
+Do not teach until Start Learning. For `optional` or `skip`, finish normally;
+`/codecall` always remains available.

package/skill/references/trigger-policy.md

@@ -0,0 +1,42 @@
+# Automatic learning recommendation policy
+
+Use this policy only after an implementation is complete. It controls a
+non-blocking recommendation, never automatic teaching.
+
+1. Confirm inspectable evidence: the completed task plus relevant conversation
+   facts or changed-file evidence. If unavailable, skip automatic prompting.
+2. Identify signals from the implementation, not line count.
+3. Recommend only with one strong signal or two distinct moderate signals.
+4. State the evidence-backed reason and affected file or conversation fact.
+5. Remember the evaluated concept cluster for this agent session. Do not show a
+   second card for the same cluster. Do not persist learner history.
+
+## Strong signals
+
+- New external dependency, integration, protocol, schema, or public interface.
+- New architecture boundary or cross-component execution path.
+- Security, authentication, authorization, concurrency, reliability,
+  data-consistency, or failure-handling behavior.
+- An explicit non-obvious tradeoff or consequential design decision.
+
+## Moderate signals
+
+- New reusable pattern, framework capability, algorithm, lifecycle, or state
+  model.
+- Connected components with a dependency relationship.
+- New configuration, observability, validation, testing strategy, or deployment
+  behavior that changes system operation.
+
+## Exclusions and outcomes
+
+Skip automatic prompting for formatting, copy, comments, mechanical renames,
+generated-file-only changes, isolated dependency bumps, and straightforward
+localized fixes. Skip test-only work unless it introduces a transferable test
+strategy, harness, boundary, or behavior.
+
+- `recommend`: show Start Learning / Skip and wait for consent.
+- `optional`: do not interrupt; leave `$codecall` and `/codecall` available.
+- `skip`: do nothing automatically.
+
+Manual invocation always remains available. Never request an API key or invoke
+an external learning model.