Repository navigation
[Draft]feat(agent): add bounded episodic recall - #122
Closed
vinceblock99 wants to merge 1 commit into
Closed
vinceblock99 wants to merge 1 commit into
vinceblock99 wants to merge 1 commit into
Conversation
Contributor
Author
|
Closing: bounded episodic recall is overkill relative to current needs. The value depends on unmet prerequisites (semantic search, single-machine history accumulation, and a turn-numbering bug fix), while its cost (a hand-written format_v3 parser in the CLI layer, added prompt-injection surface, and a blurred Memory-vs-evidence boundary) is immediate. Shelving in favor of the human-approved Memory flow. Branch preserved; can be reopened if we later confirm real demand for episodic recall. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
Vault Memory intentionally contains only reviewed, durable knowledge. When no Memory exists yet—or the available Memory is incomplete—an agent has no safe way to find relevant work from earlier local sessions. That can cause repeated investigation, but treating raw transcripts as Memory would bypass the human approval boundary.
What changed
Adds a read-only command:
atomic agent recall "authentication timeout" --jsonIt searches prior agent changes referenced by local session state and visible from the current view or its parent chain. It ranks saved reasoning, change metadata, file paths, prompts, and any already-persisted condensed turn text, then returns a small evidence bundle with the exact session, turn, and change hash.
Results are explicitly marked as historical evidence—not durable Memory or instructions. The command does not call an LLM, write to the Vault, promote a transcript, or create RDF Memory lineage.
The reader also enforces file, section, decompression, text-size, and collection-count limits. It excludes sibling drafts, unrelated local changes, imported teammate episodes, and redacted turn content.
Why
This gives agents a fallback after
atomic vault context: use approved Memory first, then inspect previous work only when needed. It helps avoid repeating research without weakening the reviewed Memory workflow.How to test
Expected: each result says it is non-durable evidence and includes
session_id,turn_number, and the fullchange_hash. An unrelated query returns an empty result set.