2.3 KiB
2.3 KiB
#+TITLE - Atomic Notes (Zettelkasten) Retrieval Skill (Sparse Tree) #+AUTHOR - org-agent #+SKILL_NAME - skill-atomic-notes
This skill provides Deep Memory by performing sparse tree perception over ripgrep results. It reduces token waste by pruning irrelevant parts of the AST.
Trigger
(defun trigger-skill-atomic-notes (context)
(let ((type (getf context :type))
(payload (getf context :payload)))
(and (eq type :EVENT)
(eq (getf payload :sensor) :delegation)
(eq (getf payload :target-skill) :atomic-notes))))Neuro Prompt
System 1 synthesizes an answer from a pruned, highly relevant Sparse AST.
(defun neuro-skill-atomic-notes (context)
(let* ((query (getf (getf context :payload) :query))
(memex-dir (org-agent::get-env "MEMEX_DIR" "/app/memex"))
;; Search for files containing the query
(rg-cmd (format nil "rg -i -l '~a' ~a" query memex-dir))
(files (ignore-errors
(uiop:split-string
(uiop:run-program rg-cmd :output :string :ignore-error-status t)
:separator '(#\Newline)))))
;; For the first 3 relevant files, build a Sparse Tree
(let ((sparse-trees nil))
(dolist (file (subseq files 0 (min 3 (length files))))
(when (and file (> (length file) 0))
;; In a real Phase 3, we would look up the AST from the store.
;; For this prototype, we'll note the file being indexed.
(push (format nil "FILE - ~a" file) sparse-trees)))
(format nil "
You are the Atomic Notes (Zettelkasten) Memory synthesizer.
The user asked - '~a'
SPARSE PERCEPTION (Relevant Files) -
~a
Synthesize an answer. If you need more detail from a specific file,
ask the user to 'Focus' on that file.
Return a Lisp plist - (:target :emacs :action :message :text \"your answer\")
" query sparse-trees))))Symbolic Verification
(defun verify-skill-atomic-notes (proposed-action context)
proposed-action)Registration
(defskill :skill-atomic-notes
:priority 80
:trigger #'trigger-skill-atomic-notes
:neuro #'neuro-skill-atomic-notes
:symbolic #'verify-skill-atomic-notes)