amr/memex
1
0
Fork 0
Code Issues 12 Pull Requests Actions Packages Projects 1 Releases Wiki Activity

docs: mark v0.2.0 Autonomous Self-Editing Foundation tasks as DONE

Browse Source 
- Exposed Structural AST Editing Tools.
- Implemented Reflection Loops.
- Hardened Actuators (path-traversal and Merkle snapshots).
- Implemented tool permission tiers (ask/allow/deny).
- Validated skill hot-reload capability.
This commit is contained in:
Amr Gharbeia
2026-04-27 13:44:50 -04:00
parent 5e68db2464
commit bdb4b30012
1 changed files with 6 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
gtd.org
View File

@@ -110,7 +110,7 @@ Elevating the user interface from raw shell piping to a high-fidelity, native Li
Roadmap basis: Evolutionary roadmap from README.org. Working backwards from SOTA parity.
**** TODO 0. Autonomous Self-Editing Foundation
**** DONE 0. Autonomous Self-Editing Foundation
***** DONE org-skill-lisp-repair (Lisp syntax repair)
- Deterministic: auto-balance parens via paren-counting
- Probabilistic: LLM generates surgical fix on =:syntax-error= events
@@ -121,19 +121,19 @@ Roadmap basis: Evolutionary roadmap from README.org. Working backwards from SOTA
- Create/update/delete headlines, set properties, manage TODO states
- Handle =id:= links and internal links
- Pure Lisp implementation (no Emacs subprocess)
***** TODO Expose Structural AST Editing Tools
***** DONE Expose Structural AST Editing Tools
- Wrap org-skill-emacs-edit into def-cognitive-tool definitions
- Force LLM to use semantic node updates instead of regex file I/O
***** TODO Implement Reflection Loops
***** DONE Implement Reflection Loops
- Feed rejection traces (syntax errors, policy blocks) back to LLM to trigger self-correction
***** TODO Harden Actuators
***** DONE Harden Actuators
- Fix path-traversal vulnerabilities in file I/O tools (e.g. :write-file)
- Enforce Merkle-snapshots on all state-modifying actions
***** TODO Implement tool permission tiers (ask/allow/deny)
***** DONE Implement tool permission tiers (ask/allow/deny)
- Per-tool permission plist stored in org-object
- =generate-tool-belt-prompt= filters denied tools before LLM sees them
- Ask-tier prompts user before execution
***** TODO Implement skill hot-reload (=:reload-skill= tool)
***** DONE Implement skill hot-reload (=:reload-skill= tool)
- Swap compiled skill files without breaking active sockets
- Reload skill into jailed package namespace
- DONE: Added :reload-skill, :read-file, :write-file, :replace-string tools