abstractskill 0.3.4__tar.gz → 0.3.6__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {abstractskill-0.3.4/src/abstractskill.egg-info → abstractskill-0.3.6}/PKG-INFO +4 -4
- {abstractskill-0.3.4 → abstractskill-0.3.6}/README.md +3 -3
- {abstractskill-0.3.4 → abstractskill-0.3.6}/pyproject.toml +1 -1
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/__init__.py +1 -1
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/catalog.yaml +1 -1
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/abstract-release/SKILL.md +45 -4
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/abstract-release/references/release-process.md +65 -6
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/validations.yaml +2 -2
- {abstractskill-0.3.4 → abstractskill-0.3.6/src/abstractskill.egg-info}/PKG-INFO +4 -4
- {abstractskill-0.3.4 → abstractskill-0.3.6}/tests/test_bundled.py +2 -2
- {abstractskill-0.3.4 → abstractskill-0.3.6}/LICENSE +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/setup.cfg +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/bundled.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/catalog.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/demand.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/errors.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/hash.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/loader.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/models.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/parser.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/policy.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/prompt.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/advisories.yaml +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/guidance.yaml +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/licenses/meshvault-live-editing.LICENSE +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/licenses/verification-before-completion.LICENSE +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/abstract-release/agents/openai.yaml +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/abstract-release/references/package-discovery.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/abstract-release/references/package-inventory.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/abstract-release/references/readiness-audit.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/abstract-release/scripts/abstract_release_plan.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/abstract-release/scripts/pypi_visibility_gate.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/abstractframework-gateway/SKILL.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/abstractframework-website/SKILL.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/abstractframework-website/references/check_site.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/abstractframework-website/references/preview_server.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/adr/SKILL.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/adr/agents/openai.yaml +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/adr/references/adr-template-and-workflow.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/adr/references/default-adr-catalog.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/adr/references/reader-first-adr-style.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/adversarial-iteration/SKILL.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/agora-collaboration/SKILL.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/agora-collaboration/references/failure-ledger.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/agora-collaboration/references/mechanics.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/architect/SKILL.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/architect/agents/openai.yaml +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/architect/references/reviewer-memory.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/backlog/SKILL.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/backlog/agents/openai.yaml +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/backlog/references/abstractframework-board.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/backlog/references/hub-work-join.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/backlog/references/layout-and-templates.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/backlog/references/maintenance-checklists.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/cicd/SKILL.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/cicd/agents/openai.yaml +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/cicd/references/docs-and-containers.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/cicd/references/maintenance-playbook.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/cicd/references/package-publishing.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/cicd/references/repository-setup.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/cicd/references/workflow-patterns.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/coredoc/SKILL.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/coredoc/agents/openai.yaml +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/coredoc/references/agent-skill.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/coredoc/references/core-doc-set.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/coredoc/references/llms-files.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/coredoc/references/maintenance-playbook.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/entity-observation/SKILL.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/entity-self-knowledge/SKILL.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/entity-self-knowledge/references/capability_map.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/meshvault-live-editing/SKILL.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/review/SKILL.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/review/agents/openai.yaml +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/review/references/reviewer-memory.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/uxreview/SKILL.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/uxreview/agents/openai.yaml +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/uxreview/references/reviewer-memory.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/verification-before-completion/SKILL.md +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/selection.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/tree.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/trust.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/validation.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill.egg-info/SOURCES.txt +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill.egg-info/dependency_links.txt +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill.egg-info/requires.txt +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill.egg-info/top_level.txt +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/tests/test_audience.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/tests/test_catalog.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/tests/test_demand.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/tests/test_loader.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/tests/test_parser.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/tests/test_policy.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/tests/test_refresh_shelf_pins.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/tests/test_selection.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/tests/test_shelf.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/tests/test_tree.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/tests/test_trust.py +0 -0
- {abstractskill-0.3.4 → abstractskill-0.3.6}/tests/test_vendor_script.py +0 -0
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: abstractskill
|
|
3
|
-
Version: 0.3.
|
|
3
|
+
Version: 0.3.6
|
|
4
4
|
Summary: Agent Skills (SKILL.md) parsing, discovery, trust gating, and demand derivation for AbstractFramework
|
|
5
5
|
Author-email: Laurent-Philippe Albou <contact@abstractcore.ai>
|
|
6
6
|
License-Expression: MIT
|
|
@@ -61,9 +61,9 @@ pip install abstractskill
|
|
|
61
61
|
|
|
62
62
|
The wheel carries the library and the curated skill registry (the shelf below).
|
|
63
63
|
|
|
64
|
-
What's new in 0.3.
|
|
65
|
-
|
|
66
|
-
the [changelog](CHANGELOG.md#
|
|
64
|
+
What's new in 0.3.5 (0.3.6 corrects the gate's flags): `abstract-release` makes CI green before anything is
|
|
65
|
+
pushed (a pre-push gate, no PR red by design, red main as an incident, a
|
|
66
|
+
flaky-test policy); see the [changelog](CHANGELOG.md#035---2026-11-03).
|
|
67
67
|
|
|
68
68
|
## The bundled skill registry
|
|
69
69
|
|
|
@@ -28,9 +28,9 @@ pip install abstractskill
|
|
|
28
28
|
|
|
29
29
|
The wheel carries the library and the curated skill registry (the shelf below).
|
|
30
30
|
|
|
31
|
-
What's new in 0.3.
|
|
32
|
-
|
|
33
|
-
the [changelog](CHANGELOG.md#
|
|
31
|
+
What's new in 0.3.5 (0.3.6 corrects the gate's flags): `abstract-release` makes CI green before anything is
|
|
32
|
+
pushed (a pre-push gate, no PR red by design, red main as an incident, a
|
|
33
|
+
flaky-test policy); see the [changelog](CHANGELOG.md#035---2026-11-03).
|
|
34
34
|
|
|
35
35
|
## The bundled skill registry
|
|
36
36
|
|
|
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
|
|
|
4
4
|
|
|
5
5
|
[project]
|
|
6
6
|
name = "abstractskill"
|
|
7
|
-
version = "0.3.
|
|
7
|
+
version = "0.3.6"
|
|
8
8
|
description = "Agent Skills (SKILL.md) parsing, discovery, trust gating, and demand derivation for AbstractFramework"
|
|
9
9
|
readme = "README.md"
|
|
10
10
|
requires-python = ">=3.10"
|
|
@@ -16,7 +16,7 @@
|
|
|
16
16
|
# four yaml files shipped in the abstractskill wheel). Bump it whenever any
|
|
17
17
|
# of that content changes; seed_registry() records it in <dest>/.seeded.json
|
|
18
18
|
# and tests/test_bundled.py pins it to a content digest.
|
|
19
|
-
version: "2026.11.
|
|
19
|
+
version: "2026.11.04"
|
|
20
20
|
|
|
21
21
|
skills:
|
|
22
22
|
|
|
@@ -48,6 +48,40 @@ missing-package discovery and the distinction between a release blocker and late
|
|
|
48
48
|
4. A worker that cannot run `git merge` or `git rebase` or `git switch` asks the coordinator to do it
|
|
49
49
|
immediately; it never leaves the branch unmerged and moves on.
|
|
50
50
|
|
|
51
|
+
## CI is green before it is pushed (AbstractFramework ADR 0050)
|
|
52
|
+
|
|
53
|
+
Why: on 2026-11-03, 22 CI runs failed in 24 hours. Ten were red by design: a branch that needed an
|
|
54
|
+
unpublished sibling, pushed two or three times anyway. Seven were red MAINs, all of them flaky timing
|
|
55
|
+
tests. The operator got ten failure e-mails overnight. A CI failure is the exception, never the normal path.
|
|
56
|
+
|
|
57
|
+
1. **The gate runs before ANY push.** Run root `scripts/ci_local.sh <repo>` (`docs/ci-local.md`). It
|
|
58
|
+
merges the commit into a freshly fetched `origin/main`, runs every job of the repo's workflows
|
|
59
|
+
(as the push to main where a workflow has one, so main's full matrix), each in a fresh environment
|
|
60
|
+
installed from the registries, and exits 0 only when every runnable job passed. Jobs it cannot run
|
|
61
|
+
here (Windows, Linux sandbox) are printed as SKIPPED and fail the gate unless `--allow-skipped`. Where
|
|
62
|
+
the gate is absent, run the workflows' job commands yourself: every step of every job (build,
|
|
63
|
+
lint, type-check, test, pack, not only the tests), the main-push Python matrix (main often tests
|
|
64
|
+
more versions than a PR), a fresh venv per version, scratch `HOME`, `pytest --basetemp=<scratch>`.
|
|
65
|
+
Push only on green. Put the commands and their results in the PR body.
|
|
66
|
+
2. **A PR is never red by design.** A branch whose CI needs a version not yet on its registry (a
|
|
67
|
+
sibling symbol, a raised floor, `pip install "x>=N"`) stays LOCAL until
|
|
68
|
+
`pypi_visibility_gate.py` says that version is visible. No PR is opened and nothing is pushed
|
|
69
|
+
until then. A branch whose last run is red is pushed again only with the fix for that failure
|
|
70
|
+
(`gh run list --branch <b> -L 1` first).
|
|
71
|
+
3. **Release branches are tested as CI installs them.** The gate builds every job's environment
|
|
72
|
+
fresh from the registries, so siblings come at their published versions, never from editable
|
|
73
|
+
checkouts, and a branch that needs an unpublished sibling fails it. In the manual fallback, use a
|
|
74
|
+
fresh venv the same way.
|
|
75
|
+
4. **Release PRs also run the release workflow's own validate / build / sdist / wheel-test steps
|
|
76
|
+
locally** before the PR (`scripts/ci_local.sh <repo> --release-tag vX.Y.Z`), because those jobs
|
|
77
|
+
run only on the tag, when it is too late.
|
|
78
|
+
5. **A red main is an incident.** Whoever merged fixes or reverts it within the hour and tells the
|
|
79
|
+
operator at once. Re-running until green is not a fix.
|
|
80
|
+
6. **Flaky tests.** No wall-clock assertion runs by default: timing bounds are opt-in (a
|
|
81
|
+
`*_TIMING_STRICT=1` switch). A test of a real server waits until the server answers HTTP, never
|
|
82
|
+
for a log line. A test that flaked once is fixed or quarantined, with a backlog item, before the
|
|
83
|
+
next merge.
|
|
84
|
+
|
|
51
85
|
## Before you start
|
|
52
86
|
|
|
53
87
|
- Default to a read-only plan. Commits, pushes, tags, workflow dispatches, publishes, CI reruns and
|
|
@@ -125,14 +159,17 @@ Repeat for each package in the plan's `release_wave`, lowest order first.
|
|
|
125
159
|
`Unreleased` into one dated `CHANGELOG.md` section for the version. Run `coredoc`'s release
|
|
126
160
|
checklist: version source, README badges and install snippets, package metadata, llms files
|
|
127
161
|
regenerated with the package's generator and committed, backlog/ADR trace.
|
|
128
|
-
3. **Local proof.** Run the
|
|
162
|
+
3. **Local proof.** Run the pre-push gate (`scripts/ci_local.sh <repo> --release-tag vX.Y.Z`, see "CI
|
|
163
|
+
is green before it is pushed"). Then the package tests. For anything that touches an engine, run a real-weights
|
|
129
164
|
end-to-end check on the machine, not a mock. For gateway routes, go through a scratch gateway
|
|
130
165
|
over HTTP (recipe in `references/release-process.md`). A local failure can be the stale venv,
|
|
131
166
|
not the code: CI installs from the published pins, so decide per failure whether it is a code
|
|
132
167
|
bug or the environment (keys in the shell, CLIs on PATH, missing extras, older siblings).
|
|
133
168
|
4. **Adversarial review before merge.** A read-only reviewer that did not write the diff reviews
|
|
134
169
|
it. Fix the confirmed findings, re-verify, and record what was found.
|
|
135
|
-
5. **Pull request, merge only on all-green.**
|
|
170
|
+
5. **Pull request, merge only on all-green.** Open it only after the gate passed, and only when every
|
|
171
|
+
version its CI installs is already on its registry: otherwise the branch stays local. The PR body
|
|
172
|
+
lists the gate's commands and results.
|
|
136
173
|
|
|
137
174
|
```bash
|
|
138
175
|
gh pr create --base main --head <branch> --title "<package> X.Y.Z: <summary>" --body-file <notes.md>
|
|
@@ -183,7 +220,9 @@ python3 "$SKILL_DIR/scripts/pypi_visibility_gate.py" --wait 900 npm:@abstractfra
|
|
|
183
220
|
python3 "$SKILL_DIR/scripts/pypi_visibility_gate.py" --wait 900 crates:abstractgateway-console==X.Y.Z
|
|
184
221
|
```
|
|
185
222
|
|
|
186
|
-
8. **Floor propagation.**
|
|
223
|
+
8. **Floor propagation.** Only after the visibility gate passed (a dependent's branch with the new floor
|
|
224
|
+
stays local until then: its gate installs the new floor from the registry), raise the floor in each direct
|
|
225
|
+
dependent the plan lists under
|
|
187
226
|
`dependency_floor_reviews`. Update the tests that pin floors literally in the same change
|
|
188
227
|
(for example AbstractGateway's `tests/test_gateway_install_profiles.py` and
|
|
189
228
|
`live_deltas.ABSTRACTRUNTIME_FLOOR`), respect the cross-package floor rules in
|
|
@@ -192,6 +231,8 @@ python3 "$SKILL_DIR/scripts/pypi_visibility_gate.py" --wait 900 crates:abstractg
|
|
|
192
231
|
|
|
193
232
|
## Gates (never skipped without the operator's approval)
|
|
194
233
|
|
|
234
|
+
0. The pre-push gate is green before every push; no PR is red by design; a red main is fixed or
|
|
235
|
+
reverted within the hour (AbstractFramework ADR 0050).
|
|
195
236
|
1. Adversarial review of every diff before merge.
|
|
196
237
|
2. Real-weights (engines) or real-gateway (routes) end-to-end proof on the machine.
|
|
197
238
|
3. Merge only when every check is `pass`; check the PR state after merging.
|
|
@@ -207,7 +248,7 @@ python3 "$SKILL_DIR/scripts/pypi_visibility_gate.py" --wait 900 crates:abstractg
|
|
|
207
248
|
`pip install abstractframework[...]` still resolves the older matrix.
|
|
208
249
|
|
|
209
250
|
Stop and report on: an ambiguous version source, a duplicate target version, an unclear publish
|
|
210
|
-
destination, a red or unknown check, a version not visible on its registry, or a floor that still
|
|
251
|
+
destination, a red pre-push gate, a red main, a red or unknown check, a version not visible on its registry, or a floor that still
|
|
211
252
|
points below a just-released version.
|
|
212
253
|
|
|
213
254
|
## Final report
|
|
@@ -70,14 +70,18 @@ For each package of the wave, lowest first:
|
|
|
70
70
|
4. **Docs.** Run `coredoc`'s release checklist: version source, README (version and registry
|
|
71
71
|
badges, install snippet, what's-new pointer), package metadata (pyproject `[project]`,
|
|
72
72
|
package.json), llms files regenerated with the package's generator and committed, docs index.
|
|
73
|
-
5. **Tests.** Run the
|
|
73
|
+
5. **Tests.** Run the pre-push gate (section 10) with `--release-tag vX.Y.Z`, which also runs the
|
|
74
|
+
release workflow's validate, build, sdist and wheel-test jobs. Then run the package's suite and the
|
|
75
|
+
end-to-end proof (section 4).
|
|
74
76
|
6. **Adversarial review.** Before merge, a reviewer that did not write the change reads the diff
|
|
75
77
|
read-only and looks for what breaks: isolation between users or accounts, double work (weights
|
|
76
78
|
loaded twice), overwritten routes, limits that disagree with the docs, missing skip guards for
|
|
77
79
|
CI. Fix confirmed findings, re-run the proofs, note the findings in the PR. Findings this caught
|
|
78
80
|
in practice: one account seeing another account's voices, weights loaded twice, a route
|
|
79
81
|
overwritten by a later registration, a 10 MB cap in code against 25 MB in the docs.
|
|
80
|
-
7. **PR and CI.** Open the PR
|
|
82
|
+
7. **PR and CI.** Open the PR only when the gate is green and every version its CI installs is on its
|
|
83
|
+
registry (a branch that needs an unpublished version stays local); put the gate's commands and
|
|
84
|
+
results in the body. Wait for checks; merge only when the guard prints exactly `pass`:
|
|
81
85
|
`gh pr checks <N> | awk -F'\t' '{print $2}' | sort -u`. A `skipping` line needs a reason
|
|
82
86
|
(a path filter, a platform job); `fail` or `pending` means stop.
|
|
83
87
|
8. **Merge.** Run `gh pr merge <N> --merge` as its own command. Never chain it after or before
|
|
@@ -102,9 +106,11 @@ For each package of the wave, lowest first:
|
|
|
102
106
|
11. **Visibility.** `pypi_visibility_gate.py --wait 900 <name>==X.Y.Z` (PyPI) or
|
|
103
107
|
`npm:@abstractframework/<pkg>==X.Y.Z` (npm). The gate checks the index pip and npm resolve
|
|
104
108
|
from, not only the project page.
|
|
105
|
-
12. **Floors.**
|
|
106
|
-
old one is still right and say why), update the tests that pin floors,
|
|
107
|
-
|
|
109
|
+
12. **Floors.** Only after step 11 passed: for each dependent in `dependency_floor_reviews`, raise the
|
|
110
|
+
floor (or confirm the old one is still right and say why), update the tests that pin floors,
|
|
111
|
+
refresh lockfiles, run the dependent's gate (it installs the new floor from the registry, so it
|
|
112
|
+
passes only after step 11 did), rerun the planner, continue.
|
|
113
|
+
A dependent prepared earlier waits as a local branch, unpushed.
|
|
108
114
|
|
|
109
115
|
Do not publish a higher package until every lower version it needs is visible on its registry and
|
|
110
116
|
the lower package's release run is green for the released commit.
|
|
@@ -194,7 +200,9 @@ Stop the wave and report the blocker when:
|
|
|
194
200
|
and the report would claim users get the change;
|
|
195
201
|
- package metadata, changelog, docs or release notes disagree;
|
|
196
202
|
- the release workflow, its environment policy, or the publish destination is unclear;
|
|
197
|
-
- the version source is unclear or the target version already exists
|
|
203
|
+
- the version source is unclear or the target version already exists;
|
|
204
|
+
- the pre-push gate is red, or a branch needs a version that is not on its registry yet (keep it local);
|
|
205
|
+
- a main is red (section 10: fix or revert within the hour, tell the operator).
|
|
198
206
|
|
|
199
207
|
## 9. Pitfalls seen in practice
|
|
200
208
|
|
|
@@ -225,3 +233,54 @@ Stop the wave and report the blocker when:
|
|
|
225
233
|
merge fails (a tag, a push, a dispatch). Run it alone and check the PR state afterwards.
|
|
226
234
|
- **Someone else's work.** Operator checkouts hold uncommitted work; never stage it. Use a
|
|
227
235
|
scratch worktree when you need a clean tree.
|
|
236
|
+
|
|
237
|
+
## 10. The pre-push gate and CI hygiene (AbstractFramework ADR 0050)
|
|
238
|
+
|
|
239
|
+
Why: on 2026-11-03, 22 runs failed in 24 hours. 10 were red by design: Gateway PRs importing a Core or
|
|
240
|
+
Runtime symbol not yet on PyPI, and a Flow release branch whose smokes need `abstractruntime>=0.17.0`,
|
|
241
|
+
each pushed 2 or 3 times. 7 were red MAINs, all flaky timing tests. 14 failures would have shown locally.
|
|
242
|
+
The gate does not catch those timing flakes (all 7 red mains): only the flaky-test policy below removes them
|
|
243
|
+
(no wall-clock bound by default, HTTP readiness, fix or quarantine), once the 8 tests are rewritten: the Core
|
|
244
|
+
GC bound (2), the gateway 0966 race family (5) and the Assistant playhead (root backlog 1116).
|
|
245
|
+
|
|
246
|
+
**The gate.** `scripts/ci_local.sh <repo> [--job <name>] [--allow-skipped] [--release-tag vX.Y.Z] [--no-merge]`
|
|
247
|
+
at the root (`docs/ci-local.md`).
|
|
248
|
+
- It merges the commit into a freshly fetched `origin/main`, as GitHub merges a pull request.
|
|
249
|
+
`--no-merge` tests the commit itself.
|
|
250
|
+
- It runs every workflow under the strictest event that workflow has (the push to main where there is
|
|
251
|
+
one, so main's full matrix).
|
|
252
|
+
- Each job runs in a fresh environment whose packages come from the registries, never from the
|
|
253
|
+
editable sibling checkouts. A branch that needs an unpublished version therefore fails here, as it
|
|
254
|
+
would in CI.
|
|
255
|
+
- `--release-tag vX.Y.Z` also runs the jobs that pushing the tag would run (build, verify).
|
|
256
|
+
- Exit codes: 0 only when every runnable job passed. 1 for a failure, a commit that does not merge,
|
|
257
|
+
or a skipped job (Windows, the Linux sandbox) without `--allow-skipped`. 2 for a usage error.
|
|
258
|
+
|
|
259
|
+
**Where the gate is absent,** run what the workflows run, by hand, from `.github/workflows/*.yml`:
|
|
260
|
+
- every job and every step (`check:lock`, `npm ci`, `build`, `lint`, `test`, `pack --dry-run`; a type
|
|
261
|
+
error shows only in `build`, not in `npm test`);
|
|
262
|
+
- the push-to-main matrix, not the PR matrix (AbstractGateway PRs test 3.12; main tests 3.10-3.13);
|
|
263
|
+
- a fresh venv per Python version, installed with the job's own `pip install` line, no editable
|
|
264
|
+
sibling; scratch `HOME`, `TMPDIR`, `HF_HOME`; `pytest --basetemp=<scratch>`; no admin token, no
|
|
265
|
+
`:8080` in the environment;
|
|
266
|
+
- for a release PR, the release workflow's validate / build / sdist / test-the-wheel steps (the
|
|
267
|
+
Avatar sdist test missed `catalog.json` and failed only on the tag).
|
|
268
|
+
Put the commands and their results in the PR body.
|
|
269
|
+
|
|
270
|
+
**Never red by design.** Before opening a PR, read the versions its CI installs. If one is not on its
|
|
271
|
+
registry, the branch stays local; check with `pypi_visibility_gate.py` and push once it passes. Before
|
|
272
|
+
pushing to a branch whose last run is red (`gh run list --branch <b> -L 1`), the push must carry the
|
|
273
|
+
fix for that failure.
|
|
274
|
+
|
|
275
|
+
**Red main is an incident.** The merger fixes or reverts within the hour and tells the operator. A
|
|
276
|
+
re-run that happens to pass is not a fix. On a main whose matrix is wider than the PR's, a flake
|
|
277
|
+
that hits 1 shard in N on a PR hits about 4 in N on main.
|
|
278
|
+
|
|
279
|
+
**Flaky tests.**
|
|
280
|
+
- No wall-clock assertion by default: timing bounds run only under an opt-in switch
|
|
281
|
+
(`ABSTRACTGATEWAY_TIMING_STRICT=1` is the model).
|
|
282
|
+
- Prefer an ordering or event check to elapsed seconds.
|
|
283
|
+
- A real-server test waits until the server answers HTTP (`/api/health`), never for a log line
|
|
284
|
+
written before the port is bound.
|
|
285
|
+
- A test that flakes once is fixed, or quarantined with a backlog item, before the next merge.
|
|
286
|
+
|
|
@@ -8,12 +8,12 @@
|
|
|
8
8
|
validations:
|
|
9
9
|
- name: abstract-release
|
|
10
10
|
source: codex-skills (maintainer)
|
|
11
|
-
tree_hash:
|
|
11
|
+
tree_hash: ee486f985a7874c6e409257fe37e0db2d3b1299ac92ba15274c3ea9b2f2c46d4
|
|
12
12
|
level: adopted
|
|
13
13
|
method: first-party-adoption
|
|
14
14
|
audience: host
|
|
15
15
|
validated_by: skill
|
|
16
|
-
validated_at: '2026-11-
|
|
16
|
+
validated_at: '2026-11-03'
|
|
17
17
|
evidence:
|
|
18
18
|
parser_roundtrip: true
|
|
19
19
|
has_scripts: true
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: abstractskill
|
|
3
|
-
Version: 0.3.
|
|
3
|
+
Version: 0.3.6
|
|
4
4
|
Summary: Agent Skills (SKILL.md) parsing, discovery, trust gating, and demand derivation for AbstractFramework
|
|
5
5
|
Author-email: Laurent-Philippe Albou <contact@abstractcore.ai>
|
|
6
6
|
License-Expression: MIT
|
|
@@ -61,9 +61,9 @@ pip install abstractskill
|
|
|
61
61
|
|
|
62
62
|
The wheel carries the library and the curated skill registry (the shelf below).
|
|
63
63
|
|
|
64
|
-
What's new in 0.3.
|
|
65
|
-
|
|
66
|
-
the [changelog](CHANGELOG.md#
|
|
64
|
+
What's new in 0.3.5 (0.3.6 corrects the gate's flags): `abstract-release` makes CI green before anything is
|
|
65
|
+
pushed (a pre-push gate, no PR red by design, red main as an incident, a
|
|
66
|
+
flaky-test policy); see the [changelog](CHANGELOG.md#035---2026-11-03).
|
|
67
67
|
|
|
68
68
|
## The bundled skill registry
|
|
69
69
|
|
|
@@ -38,8 +38,8 @@ LICENSES = (
|
|
|
38
38
|
# The bundle version must move whenever the bundled content moves. When this
|
|
39
39
|
# test fails: bump `version:` in src/abstractskill/registry/catalog.yaml, then
|
|
40
40
|
# update BOTH values below (the digest printed in the failure message).
|
|
41
|
-
PINNED_VERSION = "2026.11.
|
|
42
|
-
PINNED_DIGEST = "
|
|
41
|
+
PINNED_VERSION = "2026.11.04"
|
|
42
|
+
PINNED_DIGEST = "f6aab0b88c4a94824ea2881d65d010b4331d46d7d1d03cd8b37f45af8b5c8294"
|
|
43
43
|
|
|
44
44
|
|
|
45
45
|
def _bundle_digest(root: Path) -> str:
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
{abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/adr/agents/openai.yaml
RENAMED
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
{abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/architect/SKILL.md
RENAMED
|
File without changes
|
|
File without changes
|
|
File without changes
|
{abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/backlog/SKILL.md
RENAMED
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
{abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/coredoc/SKILL.md
RENAMED
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
{abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/review/SKILL.md
RENAMED
|
File without changes
|
|
File without changes
|
|
File without changes
|
{abstractskill-0.3.4 → abstractskill-0.3.6}/src/abstractskill/registry/skills/uxreview/SKILL.md
RENAMED
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|