fix(converter): track upstream v3.8.4 (.md agent rename) + refresh idle figures#30
Merged
Conversation
Upstream v3.8.4 renamed agent files from `<name>.agent.md` to plain `<name>.md` (the agents/ dir disambiguates now, not the extension). extract.py globbed `*.agent.md`, found zero agents, and produced an empty manifest -- so publish-dist has failed every run since v3.8.4 was tagged, leaving ce-lite stuck on v3.8.3 with auto-tracking silently broken for ~5 days. - extract.py: discover `agents/*.md`; accept a filename under either `.agent.md` (<=v3.8.3) or `.md` (v3.8.4+). Dist output is unchanged (already `<name>.md`, derived from the frontmatter name). - validate.py: round-trip resolves the upstream file under either extension. - tests: add extract_agents discovery coverage (plain `.md`, legacy `.agent.md`, and the filename/name-mismatch guard) -- the gap that let this regress with no failing unit test. Verified: full pipeline green against compound-engineering-v3.8.4 (49 personas, round-trip byte-equivalent, version 3.8.4-lite).
…st overclaims Figures: the idle-saving number was scattered across six files with four different values (58.8k / 71k / 77k / 99.9k). Reconcile to one canonical, drift-resistant statement measured via interactive /context: ~2k idle tokens per registered agent, ~88k on upstream v3.8.4, net ~= -94k, with all 49 personas preserved. The char-count estimates undercounted; the /context "Custom agents" slot is the real measure of what occupies the window (and `claude plugin details` under-reports it -- it counts descriptions, not the loaded bodies). Headless: stop implying ce-lite has working `claude -p` headless tests. The integration-eval routing probe and measure-baseline.py are opt-in/unreliable, not gates -- caveat them accordingly. Left refine-keyword-rules.py (a generation tool, not a test) as-is.
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
1 participant
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.
Why
Upstream v3.8.4 renamed agent files from
<name>.agent.mdto plain<name>.md(theagents/dir disambiguates now, not the extension).extract.pyglobbed*.agent.md, found zero agents, and produced an empty manifest — sopublish-disthas failed every run since v3.8.4 was tagged (~5 days), leaving ce-lite stuck on v3.8.3 with auto-tracking silently broken.What
Converter fix:
extract.pydiscoversagents/*.md, accepting a filename under either.agent.md(≤v3.8.3) or.md(v3.8.4+). Dist output unchanged (already<name>.md, from the frontmatter name).validate.pyround-trip resolves the upstream file under either extension.extract_agentsregression tests (plain.md, legacy.agent.md, filename/name-mismatch guard) — the gap that let this regress with no failing unit test.Docs:
/contextis the real measure of window occupancy;claude plugin detailsunder-reports it.)claude -pheadless references that implied working tests (integration-eval,measure-baseline.py) — opt-in/unreliable, not gates.Verification
compound-engineering-v3.8.4: 49 personas, round-trip byte-equivalent, version3.8.4-lite.nix flake check(treefmt + actionlint + pytest) passes; 164 tests (+3 new).skill-rules.jsonis byte-identical v3.8.3 → v3.8.4 — persona routing unchanged, so noclaude -pregeneration needed.After merge
gh workflow run publish-dist --field upstream_tag=compound-engineering-v3.8.4will now succeed, bumping ce-lite to3.8.4-liteand ending the auto-tracking stall.