at-x3s sprint-start: push task-start recall into the session at claim time#51
Open
tcovert-c137[bot] wants to merge 1 commit into
Open
at-x3s sprint-start: push task-start recall into the session at claim time#51tcovert-c137[bot] wants to merge 1 commit into
tcovert-c137[bot] wants to merge 1 commit into
Conversation
… time The mu-rkhj lesson, operator-approved fix: the settled design for the exact subsystem being claimed sat in memory (top recall hit for the bead title) and a staged spec, and nothing surfaced either at task start — the prompt-keyed recall hook fires on user wording, and manual search depends on model discipline. The operator paid for the same design twice. After the claim (and workspace setup), sprint-start now runs SEMANTIC recall — agent memory recall + agent kx recall — on the bead's title (the label in --no-bead mode) and prints the top 3 hits of each to stderr under a READ-BEFORE-IMPLEMENTING banner with the ids to agent-memory-show. Semantic recall, not FTS search, on purpose: search is AND-semantics FTS5 and one unmatched title word empties it; recall tolerates raw titles (verified on the motivating case — the mu-rkhj title recalls the operator's messaging design and event-driven standard as top hits). Best-effort by construction: no agent/python3 on PATH, fetch failure, or empty results all skip silently; nothing can block or fail the sprint; stdout (the trailing cd line scripts consume) is untouched. A bare bead id embeds nothing, so if the title can't be fetched the recall is skipped rather than searched-for-garbage.
sahuagin
force-pushed
the
cc/at-x3s-sprint-start-recall
branch
from
10c1dfb to
eed1495
Compare
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.
Fixes at-x3s. The mu-rkhj lesson made structural: the settled dialogue design sat in memory (top recall hit for the bead's own title) and a staged spec, and nothing surfaced either at task start — the prompt-keyed recall hook fires on user wording, and manual search depends on model discipline. The operator paid for the same design twice.
After a successful claim (and workspace setup),
sprint-startnow runs semantic recall —agent memory recall+agent kx recall— on the bead's title (the label in--no-beadmode) and prints the top 3 hits of each to stderr under aREAD BEFORE IMPLEMENTINGbanner withagent memory show <id>pointers. Pull becomes push, keyed on the work itself rather than prompt wording.Semantic recall rather than FTS
search, deliberately: search is AND-semantics FTS5, so one unmatched title word ("rip") empties the result set; recall's embedding+cosine tolerates raw titles. Verified live on the motivating case — the mu-rkhj bead title recallsoperator-messaging-substrate-preference,operator-event-driven-systems-standard, and the epic note namingPUSH-REDESIGN.mdas the top hits, any one of which would have prevented the deviation.Best-effort by construction: missing
agent/python3, endpoint failure, unfetchable title, or empty results all skip silently; nothing can block or fail the sprint; stdout (the trailingcdline scripts consume) is untouched — verified.Gate note:
just fmt-checkandjust testgreen;just check's clippy leg is broken on main independent of this shell-only change (unknown lint: clippy::chunks_exact_to_as_chunks, toolchain skew) — filed as at-7pb.🤖 Generated with Claude Code
https://claude.ai/code/session_01REB3QZvP3arA2cLMVRgYam