v0.8.0 release: Updated CHANGELOG.md, bumped
Project.toml to 0.8.0, tagged and pushed. Created GitHub Release with
changelog notes.
CI fix: JuliaFormatter blank line after
!!! warning admonition in XAct.jl docstring.
474a4f1.
Python packages: Bumped xact-py and sxact versions
to 0.8.0 in pyproject.toml. Re-created GitHub Release (tag moved to
HEAD) to trigger PyPI publish — succeeded. xact-py 0.8.0
live on PyPI. 6b99266.
TagBot workflow: Added
.github/workflows/TagBot.yml — triggers on
push: tags: ['v*'] to auto-create GitHub Releases.
92a0280.
CompatHelper workflow: Added
.github/workflows/CompatHelper.yml — daily midnight UTC
cron for dependency version PRs. 1f17038.
Contributing docs: Updated both
CONTRIBUTING.md and docs/src/contributing.md
with release process, TagBot/CompatHelper, and
@JuliaRegistrator register instructions.
Benchmark investigation: Ran XPerm benchmarks
locally. Confirmed CI "regressions" (2×–600×) are a baseline
format mismatch — old baseline test_ids (short names) don't
match current format (full path). Not a real performance regression.
Triggered record-ci-baseline workflow.
Arc: Two release cycles completed — bug fixes,
Session/reset!() improvements, CI maintenance, release infrastructure,
and Python package publication.
Next: Verify record-ci-baseline
completes and benchmark check passes. Then register in Julia General
Registry via @JuliaRegistrator register.
Tasks created
Decisions
TagBot triggers on push: tags: ['v*'] (immediate)
rather than issue_comment from JuliaTagBot (delayed)
Python packages versioned in lockstep with Julia package
(0.8.0)
CI baseline needs re-recording on CI hardware due to test_id format
change
beads backup: use tracked issues.jsonl + backup config
rather than Dolt remote — simpler, no infra needed, issues push with
git push
Session 14:01 (testaruda — 1
session)
OpenSpec specs: Deployed SRS v0.3 as 18 capability
specs under openspec/specs/ — every requirement from ARCH
to ENGINE extracted with scenarios. All 18 validated via
openspec validate --specs --strict.
Beads issues: Created 17 issues covering
implementation gaps vs specs. Ran issue-review skill —
added acceptance criteria to all 19 issues (1 closed, folded into
features), wired dependency graph, restructured horizontal tickets into
tracer bullets (adapter protocol → 3 phases, ordering+agent → 2
issues).
Beads backup fix: Removed blanket
.beads/ ignore from ~/.gitignore. Project
.gitignore now ignores only internal beads state via
.beads/* + un-ignores issues.jsonl and
config.yaml. Enabled backup in config. Applied same fix
across all 17 bead-using repos in the charly org (atril, bichos, dont,
espectacular, fabbro, fotos, khipu, miblioteca, nayra, paranoid,
pretender, rizomas, ruta, superficies, testaruda, tRAGar, wai). All
committed and pushed.
Arc: Testaruda went from plain specs + empty beads
to a fully wired workflow: specs → espectacular contracts → beads issues
with deps and acceptance criteria → auto-backup via git push.
issue-review: Ran full Rule of 5 review on 17 beads
issues. Fixed all findings: added acceptance criteria to every issue,
wired dependency graph between issues, restructured horizontal tickets
(adapter protocol split into 3 tracer bullets, ordering+agent split into
2, catch-all test ticket closed and folded into features). 19 open
issues, 1 closed.
Beads backup debug: Investigated why
issues.jsonl wasn't pushing. Root cause:
~/.gitignore had blanket .beads/ rule + no
Dolt remote configured. Fix: removed .beads/ from global
gitignore, added .beads/* + un-ignore
issues.jsonl and config.yaml in project
.gitignore, enabled backup in
config.yaml.
Cross-repo rollout: Applied same fix to all 16
other bead-using charly repos. Fixed bichos which had duplicate
.beads/ line overriding un-ignore rules. Cleaned stale git
locks in pretender. All 17 repos now track issues.jsonl and
push with git push.
Decision: beads backup via tracked JSONL export +
backup config (no Dolt remote needed) — simpler, no infra, issues travel
with code.
Evaluation review: Reviewed
testaruda-evaluation.md — confirmed document is still
accurate (HEAD unchanged since retest round). Identified three
unaddressed follow-up items from §8.4.
Justfile fix (I6 / Backlog gap B): Replaced broken
Markdown/HTML preamble with 16 proper just recipes (build, test, clippy,
fmt, fmt-check, check, ah, init, select, ingest, explain, clean, doc,
release, test-v, default). just --list now works,
just build compiles. Filed and closed beads issue
testaruda-m2p.
Dependency link (Backlog gap A): Ran
bd link testaruda-mfx testaruda-9uw — mfx now correctly
depends on 9uw. Added note about adding an I2 regression re-test as
acceptance criterion.
Cold-start spec gap (S9): Created openspec change
proposal add-coldstart-classification — adds
TIA-CHG-009 requiring idempotent unresolved classification
for first-observed content units. Validated
(openspec validate --strict passes).
Next: Review and approve the cold-start proposal
when ready; resume P1 implementation (content fingerprinting, adapter
protocol, CI exit codes).
Session 17:36 (cositos — 1
session)
CI/docs/release infra: Created 3 GitHub Actions
workflows — verify.yml (docs-build + Julia parity
precompile), docs.yml (Quarto → GitHub Pages deploy),
release.yml (PyPI + npm + GitHub Release on
git tag v*). All now 🟢 green.
History recovery: Replaced local .git
with full 168-commit history from archive zip
(~/para/projects/work/dev/archive/cositos.zip), cleaned via
git-filter-repo (zero client-name references). Used
filter-repo to fix author info across all commits to
sashakile <[email protected]>.
Remote setup: Added
[email protected]:sashakile/cositos.git, force-pushed 169
commits (168 original + 1 infra commit). Set _quarto.yml
site-url.
CI fixes: Added explicit
julia-actions/setup-julia@v2 + parity deps precompile to
both docs-build jobs. Made coverage-audit.py gracefully
tolerate missing pretender binary (CI gap,
cositos-dad.19).
Beads: Initialized with prefix
cositos, set git config beads.role maintainer.
Created and closed 5 issues for the infra work.
Wai: Advanced phase research → design → plan →
implement → review. Captured design and plan artifacts.
Next: Publish PyPI packages when ready
(mise run bump-version 0.1.0 && git tag v0.1.0 && git push origin v0.1.0).
The @cositos/front npm package needs NPM_TOKEN
secret set before publishing.