learn

> **Manual-trigger skill.** `disable-model-invocation: true` keeps the model from > invoking this workflow automatically mid-task. When installed from this plugin, > invoke it deliberately as `/agent-docs:learn` at the end of a session. > Do not add hooks, background tasks, auto-trigger behavior, runtime storage, > vector databases, MCP integration, or external memory systems. Review what happened in this session and produce verified, reviewable knowledge proposals for the appropriate `AGENTS.md` or `docs/` surface using the exact-diff workflow below. Before classifying candidates, read and apply the shared [Knowledge Admission Policy](../../references/knowledge-admission.md). This skill produces new knowledge proposals; the policy is shared with `remember` and `curate` rather than owned by this workflow. When a candidate belongs under `docs/`, also read the [Documentation Structure Reference](../../references/doc-structure.md). ## Admission Model Non-derivability is sufficient, not necessary: - Verified, durable, non-duplicative knowledge that cannot be derived from the repository is admitted automatically and routed to the right surface. - Derivable knowledge may still be admitted when its value score justifies the target's maintenance or prompt cost. `Hidden Knowledge` remains the strict destination for non-derivable gotchas. High-value derivable commands, maps, rules, and workflows belong in their purpose-specific surfaces instead of being mislabeled as hidden ...