Repository navigation
Releases
v6.7.0
Superstacks 6.7.0
Verification, testing and planning guidance: suites nobody runs for you, independence between open changes, settings meant to last for one transaction, and generated lists kept as rows and totals.
This tag is for the owner to create later. This pull request does not push v6.7.0.
Changes
verification-before-completiongains two sections: find and run the suites that neither the pipeline nor the pre-push hook runs (real database or service, recorded output, environment flag) for every folder a change touches, and treat two open changes as independent only after searching each branch for every exported name the other changestest-driven-developmentgains a case for a setting meant to last for one transaction: test it on a pool of exactly one connection, compare the setting and the connection's identity before and after, and flip the scope by hand once to watch the test failinterrogate(rubric): a break made by hand that no test catches is a finding even when the shipped code is right, reported as a missing test with the flag or line that was flippedwriting-plansgains a section on generated lists that every open change rewrites: keep rows (one per smallest unit, regenerated by each change and compared by a test) apart from totals (regenerated on request), and regenerate rows after every merge of the main branch
v6.6.1
Superstacks 6.6.1
The recorded file hashes in the source lock now match the files, and a test keeps them matching.
This tag is for the owner to create later. This pull request does not push v6.6.1.
Changes
sources.lock.json: corrected the staledest_sha256hashes forrules/safety-overrides.mdc,skills/ticket-pipeline/SKILL.mdandskills/workflow-profile/SKILL.md; the file text did not changetests/test_lock_hashes.py: a new test that fails and names each file whose recorded hash differs or that is missing
v6.6.0
Superstacks 6.6.0
Planning, review and testing guidance for calls that change another system: record before sending, count requests and answers, and name what was proven.
This tag is for the owner to create later. This pull request does not push v6.6.0.
Changes
writing-plansgains a section on a call that changes another system: write what you are about to do, with the identifiers and the request time, before sending; decide only from what is written, counting requests sent and answers read; name each closing state by what was proven and keep the "verified" time for what the other system confirmedinterrogate(rubric): reviewers walk the lost answer, name the reply that proves each closing state, and treat an old status as evidence only when it is newer than a time stored before sendingtest-driven-developmentgains the lost-answer test shape: the fake applies the change and then fails the call with a timeout, and the test asserts the retry leaves the work open and waiting for proof
v6.5.0
Superstacks 6.5.0
Testing and review guidance: a test must be able to see the defect it names, and a reading of code is not evidence.
This tag is for the owner to create later. This pull request does not push v6.5.0.
Changes
test-driven-developmentgains a section on tests that cannot see the defect: verify a stored value through a second path (once under a non-default environment such as a non-UTC database session), assert every number a test reports, and fence the first-of-a-kind write that snapshot isolation does not protect, with a gated overlap test that always has a way to open and a time limitverification-before-completiongains a "before you say ready" step to run the whole-tree guard tests as well as the changed folder's tests, and a rule that a reading of the source is a hypothesis while a failing test is evidenceinterrogate(rubric, lead judgment): reviewers search tests for results that are logged and not asserted, check persisted values through a second path, and ask about write skew; a failing test outranks a careful reading
v6.4.0
Superstacks 6.4.0
Reviews of scanners and guards must be independent, and "ready" means every comment was read.
This tag is for the owner to create later. This pull request does not push v6.4.0.
Changes
- A change that adds or alters a finder or refuser (scanner, inventory, guard, lint rule, drift test, validator) needs a review with an independent count or trace and a set of probes, each marked found or missed. A missed form in real code today is at least high severity; one that does not exist today is recorded as a stated limit (
interrogate,code-review) verification-before-completiongains a "before you say ready" step: read every comment and review since your last one, and name the method behind none, all, safe and alwaysprgains a## Reviewsection; a body withClass: findermust name the independent check and the probes, enforced bycheck_pr_body.py
v6.3.0
Superstacks 6.3.0
Unclear assignments get plain-language questions before work starts.
This tag is for the owner to create later. This pull request does not push v6.3.0.
Changes
- New model-invoked
amaskill: look before asking, sort open points into ask, assume or look up, ask at most four questions at a time with two to four easy options each (one Recommended), record answers throughowner-decisions, then send a short brief to confirm - The quality and safety bar (test-first, verified, reviewed, secure by default) is stated as never up for a question
ama/templates/QUESTIONS.mdandama/scripts/check_questions.py(sheet validator)ticket-pipelineand the README listamafirst
v6.2.0
Superstacks 6.2.0
Merge authority comes from a record of the owner's decisions.
This tag is for the owner to create later. This pull request does not push v6.2.0.
Changes
- New model-invoked
owner-decisionsskill: find the record, read before asking, record before acting, authority from the record, createDECISIONS.mdonly on request, propose (never apply) standing rules after five like decisions DECISIONS.mdtemplate,check_decisions.pyvalidator,authority.pylookupsafety-overrides,workflow-profile,ticket-pipeline:trust_levelstays 1 by default; merge and ready authority are read from the record, and only a recorded standing rule may raise it
v6.1.1 — Superstacks
6.1.1 - 2026-10-06
Changed
- Claude marketplace
metadata.descriptionsoclaude plugin validate --strictpasses for the marketplace as well as the plugin. - Truthful wording for optional local helper scripts a skill may tell the agent to run (
git diff/git log, a user-supplied verify-loop command, an existing Playwright install). The plugin makes no network calls, collects no telemetry, and does not read credentials. - Plugin README logo uses Markdown instead of HTML.
- Codex listing: category Developer Tools, filled capabilities, three
defaultPromptentries,extensions.com.openai.publication.release_notes, and a.codexignore. - ZIP builder also emits
dist/flat/per-skill archives withSKILL.mdat the zip root.
Removed
- Committed
tests/__pycache__bytecode.
Install
| Surface | How |
|---|---|
| Claude Code | /plugin marketplace add KunanonJ/superstacks then /plugin install superstacks@superstacks |
| Cursor | Copy plugins/superstacks/ to ~/.cursor/plugins/local/superstacks/ until the public listing exists |
| Codex | codex plugin marketplace add KunanonJ/superstacks |
| claude.ai ZIP upload | python scripts/build_zips.py and upload a per-skill ZIP from dist/ (<skill>/SKILL.md) or dist/flat/ (SKILL.md at zip root) |
| ChatGPT ZIP upload | Same per-skill ZIPs, or dist/superstacks-plugin.zip for the OpenAI plugin portal |
npx skills add |
DISABLE_TELEMETRY=1 npx skills add KunanonJ/superstacks -g -s '*' --copy -y |
| Plain copy | Copy plugins/superstacks/skills/<name>/ into your agent's skills directory |
v6.1.0 — Superstacks
6.1.0 - 2026-10-06
Added
- Model-invoked
verify-loopwithverify.py,verify.sh, and a Playwright snapshot/console template that never installs Playwright. - User-invoked
mistake-to-constraint,steer-miner,intake-triage,gardening, andsample-review. - Deterministic checkers:
pr/scripts/check_pr_body.pyandticket-pipeline/scripts/pipeline_status.py. trust_levelladder (default 1) insafety-overridesandworkflow-profile.- README Philosophy section crediting the public Matt Pocock × Poteto conversation.
- A02 Crisp logo (SVG master, 512 PNG, 256/64/16 icons) and a 1280×640 GitHub social preview.
- Live GitHub URLs and install commands use
KunanonJ/superstacks. - Stacked pull request guidance in
prandticket-pipeline.
Install
| Surface | How |
|---|---|
| Claude Code | /plugin marketplace add KunanonJ/superstacks then /plugin install superstacks@superstacks |
| Cursor | Copy plugins/superstacks/ to ~/.cursor/plugins/local/superstacks/ until the public listing exists |
| Codex | codex plugin marketplace add KunanonJ/superstacks |
| claude.ai ZIP upload | python scripts/build_zips.py and upload a per-skill ZIP from dist/ |
| ChatGPT ZIP upload | Same per-skill ZIPs, or dist/superstacks-plugin.zip |
npx skills add |
DISABLE_TELEMETRY=1 npx skills add KunanonJ/superstacks -g -s '*' --copy -y |
| Plain copy | Copy plugins/superstacks/skills/<name>/ into your agent's skills directory |
v6.0.0 — Superstacks
6.0.0 - 2026-10-06
Added
- Superstacks v6 plugin at
plugins/superstacks/with Claude, Cursor, and Codex manifests. - Root marketplaces for Claude, Cursor, and Codex/local agents.
- Vendored skills from obra/superpowers, mattpocock/skills, and pstack, plus original
ticket-pipeline,workflow-profile, andsafety-overrides. - ZIP builder for claude.ai / ChatGPT skill uploads and the OpenAI plugin portal.
- Plugin linter, archive size check, and pull-request CI.
Removed
- The v5 aggregated skill catalog (archived as git tag
archive/v5-catalog).