GeezerVison

plugin-submissions (#896)

* plugin-submissions

Package the local stdio MCP server for both vendor directories and make
docs/stackql_release_process.md the source of truth for every surface.

OpenAI (#893): portable Agent Plugins layout (plugin.json with
extensions.com.openai interface, review test cases, publication notes,
onboarding skill, requires_local_executor; mcp.json stdio with
${PLUGIN_ROOT}) alongside the Codex overlay, which is now derived from
plugin.json by the renderer; 30-character short description, support and
terms URLs; vendored launcher under server/bin; make openai-plugin-zip and
build-zip.py; validate.py extended to the directory rules.

Anthropic (#895): packaging/claude-plugin plugin bundle (.claude-plugin
manifest with listing fields, inline mcpServers, README, LICENSE, icon,
skill, same vendored payload); validate-claude-plugin.py; publish-mirror.sh
and make claude-plugin-mirror push it to stackql/stackql-claude-plugin;
Anthropic submission checklist rewritten for the developer portal.

Shared: the npm launcher is now dependency-free (inline zip extraction
replaces adm-zip) and honours STACKQL_PLUGIN_CONFIG_DIR, passing --env.file
and --configfile, so both plugins vendor it verbatim; render-platforms.sh
gains the claude-plugin vector and the shared plugin payload;
skills/stackql-setup/SKILL.md is the onboarding skill for both; the
openai-plugin workflow becomes plugins.yml (both validators, both smoke
tests, render check, claude plugin validate --strict, ZIP artifact);
npm, PyPI and plugin pins bumped to the published 0.12.824.

Resolves #893
Resolves #895

Co-Authored-By: Claude Fable 5.1 <[email protected]>

* build: pin the Linux Test job to ubuntu-24.04

ubuntu-latest moved to Ubuntu 26.04 on 2026-11-04. On that image the
nginx apt-key import in the Linux Test job fails (gpg: /home/runner/.gnupg
does not exist) before any test runs, and nginx.org publishes no 26.04
suite yet. Pin the job to the image main was validated on.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

* build: correct the ubuntu-24.04 pin comment

nginx.org does publish a 26.04 (resolute) apt suite; the verified failure
on the 26.04 image is only the gpg home directory abort during the key
import, so the comment now states that alone.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

* build: pin the Docker Test jobs to ubuntu-24.04

On the Ubuntu 26.04 image that ubuntu-latest now resolves to, the newer
Docker Compose writes container progress lines (Container ... Running,
Creating, Created) into the output the mocked robot suite captures, so 33
exact-output assertions fail on the amd64 shards and the Docker Python
Test. The arm64 legs on ubuntu-24.04-arm pass. Pin the amd64 leg and the
shard verification job to the image main was validated on.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

---------

Co-authored-by: Claude Fable 5.1 <[email protected]>

omnisdk-configurable (#865)

Summary:

- Gap fill for `omnisdk`.

fast-fail-where-needed (#859)

Summary:

- Amended robot test `Page Number Pagination Missing Terminator Stops After One Page`.
- This supports fast failure for the `omnisdk` cutover testing step, which otherwise hangs for some time.

windows-multiarch-in-parallel (#841)

* windows-multiarch-in-parallel

Summary:

- Support for native windows build and test on both `amd64` and `arm64`.
- Builds run in parallel, realising time saving from prior.

* windows-multiarch-workflow-cleanup

---------

Co-authored-by: Jeffrey Aven <[email protected]>

omnisdk-query-replacement-opt-in (#797)

Summary:

- Support for limited, lightweight streamed queries via `preview` opt in.
- Pass each document-driven provider's configured stackql auth, basic included, to omnisdk, and report each method's SQL verb in SHOW METHODS.
- Added developer documentation for preview features in `docs/preview.md`.
- Added robot test `Unstable Google Kms Key Rings Joined To Crypto Keys Jsonl Row Set Matches Expectation`.
- Added robot test `Unstable Github Org Members Filtered By In List Jsonl Row Set Matches Expectation`.
- Added robot test `Unstable Github Org Update Reports Despatch`.
- Added robot test `Unstable Github Org Update Returning Jsonl Row Set Matches Expectation`.
- Added robot test `Unstable Google Kms Key Ring Insert Returning Jsonl Row Set Matches Expectation`.
- Pass each document-driven provider's configured stackql auth, basic included, to omnisdk, and report each method's SQL verb in SHOW METHODS.

mcp-go-sdk-1.8.0-and-protocol-version-pin (#796)

* mcp updates

* retire describe_resource MCP tool; discovery via list_methods -> describe_method

The MCP describe_resource tool wrapped DESCRIBE provider.service.resource,
which reports the schema of one select method resolved without WHERE
parameters, while a SELECT routes by its parameters and can land on a
method with a different shape. describe_method is the exact contract
(inputs with param_type, output fields) for the method a query routes to,
so it is now the single column source.

- Remove the tool registration, Backend.DescribeResource and its
  implementations, the interrogator SQL builder and the description file.
  The SQL DESCRIBE statement is unchanged.
- Rewrite the discovery instructions as a numbered drill-down ending in
  describe_method; sharpen the describe_method and list_methods
  descriptions; repoint dialect rules, error recovery, prompts and the
  deploy-stack resource at describe_method output fields.
- Docs: drop the tool from docs/mcp.md, the package README and the bundle
  README; the docs tool table now also lists the query library tools.
- Tests: unit test asserting the tool is not published; robot list-tools
  scenario asserts its absence and describe_method's presence; the
  describe-resource robot scenario is removed.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

---------

Co-authored-by: Claude Fable 5.1 <[email protected]>

chore: bump any-sdk and reduce ci runner cost (#790)

migrate-embedded-sqlite-to-pure-go-modernc (#786)

* modernc sqlite migration prep

* migrate-embedded-sqlite-to-pure-go-modernc (#783)

* docs(sqlfuncs): mirror CLAUDE.md as AGENTS.md for all agent sessions

Co-authored-by: jeffreyaven <[email protected]>

* deps: adopt any-sdk pure-Go sqlite backend, drop mattn/cgo fork replace

Co-authored-by: jeffreyaven <[email protected]>

* ci: retire cgo - CGO_ENABLED=0, drop C toolchains and sqlite_stackql tag, add per-artifact function smoke

Co-authored-by: jeffreyaven <[email protected]>

* docs+test: modernc backend invariants, release notes draft, robot ext-function smoke scenario

Co-authored-by: jeffreyaven <[email protected]>

* deps: bump any-sdk to alpha02 (aws_policy_equal tags parity, boolean decltype); pin sqlite_version fixture to 3.53.4

Co-authored-by: jeffreyaven <[email protected]>

* burn: remove executed migration work order

Co-authored-by: jeffreyaven <[email protected]>

---------

Co-authored-by: copilot-swe-agent[bot] <[email protected]>
Co-authored-by: jeffreyaven <[email protected]>

* finalize-pure-go-sqlite-migration-cleanup (#785)

* retire remaining cgo configuration and stale migration artifacts

Co-authored-by: jeffreyaven <[email protected]>

* preserve migration smoke-test documentation

Co-authored-by: jeffreyaven <[email protected]>

---------

Co-authored-by: copilot-swe-agent[bot] <[email protected]>
Co-authored-by: jeffreyaven <[email protected]>

* shard-postgres-docker-robot-tests

Co-authored-by: jeffreyaven <[email protected]>

* keep-stateful-robot-tests-in-one-shard

Co-authored-by: jeffreyaven <[email protected]>

* rename-release-process-doc-without-spaces

Rename `docs/stackql release process.md` to
`docs/stackql_release_process.md` per review, and update the three
references to it (two workflow header comments and the mcpb packaging
notes) so nothing points at the old name.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

* move-extension-function-smoke-to-robot

Remove the six workflow-level extension function smoke steps from
build.yml per review. The check lives in the robot scenario
`Sqlite Extension Functions Smoke Working`, which now carries the
stronger aws_policy_equal comparison the workflow steps used (Version
plus scalar-vs-array Resource), so nothing is lost in the move.

The scenario runs wherever the functional suite runs: winbuild,
linuxtest, linuxarmbuild, macosbuild, wsltest and the docker sqlite leg.
winarmsmoke and macosarmbuild have no robot harness, so the check is no
longer exercised in those two jobs.

Also drop the stale "cgo" wording from the winarmsmoke comment.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

* assert-shard-completeness-after-docker-tests

Sharding splits one robot run across several dockertest legs, so no
single leg can attest that the whole suite ran. Add a dockertestverify
job that runs once all legs conclude and fails unless, for each
(platform, db backend) group, the shards together account for every
functional test exactly once.

- Each robot-running dockertest leg uploads its output.xml; unsharded
  legs upload as shard 1 of 1.
- robot_shard.py gains a verifier. The expected tests are parsed from
  test/robot/functional with no shard modifier applied, then compared
  as a multiset against the union of the shard reports, so the total
  count and each individual test are checked independently.
- Anything that prevents verification fails it: absent shard, absent
  output.xml, inconsistent shard counts, a NOT RUN test, an expected
  group with no reports, or reports from a group that is not listed.
- dockermerge now needs dockertestverify, so an image cannot be pushed
  past a failed verification.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

* run-robot-suite-on-arm-runners

Per review: now that GitHub hosts arm64 macOS and Windows runners, run
the robot functional suite on both rather than leaving the extension
function check unexercised there.

- macosarmbuild: mirror the macosbuild robot harness (libpq, python
  deps, registry rewrite, certificates, robot run, report upload). The
  libpq path is resolved with `brew --prefix` because homebrew's prefix
  differs on Apple Silicon.
- winarmsmoke: mirror the winbuild robot harness against the amd64
  binary under x64 emulation. Python is x64 because psycopg2-binary,
  psycopg-binary and cryptography publish no win_arm64 wheels. The job
  now checks out the repo (before the artifact downloads, since checkout
  cleans the workspace) and moves from the short to the standard job
  timeout.
- winbuild: upload the MCP client as an artifact so the MCP scenarios
  run in winarmsmoke rather than being skipped.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

---------

Co-authored-by: Copilot <[email protected]>
Co-authored-by: jeffreyaven <[email protected]>
Co-authored-by: Claude Fable 5.1 <[email protected]>

modernc sqlite migration prep

reload-credentials-redesign (#761)

* reload_credentials: report installed providers, invalidate doc auth contexts, changed flag, actionable credential errors

The reload_credentials MCP tool read the lazily populated auth context
registry, so an unscoped call before any query returned no rows and a
scoped call for an installed, credentialed provider errored with
"cannot find AUTH context" (after the env re-source had already run).

The handler is now three ordered phases: re-source the configured env
file (a configured file that has gone missing is an explicit error
naming the path), invalidate auth contexts registered lazily from
provider documents (--auth contexts are retained; the cached provider
object is dropped with its context because doc auth only registers on
provider creation), then report every installed provider against the
now-current environment. Scoping is a filter; an unknown provider name
is the only scoped error. Each row carries a new `changed` field: a
sha256 comparison of the resolved credential material (successor chain
included) before and after the re-source, never emitted.

Query tools: credential resolution failures on run_select_query,
run_mutation_query and run_lifecycle_operation now name the provider
(parsed from the statement) and the recovery sequence (fix the env
file, call reload_credentials, retry). execQuery propagates statement
errors, which mutation, lifecycle and pull_provider previously
swallowed into an empty success payload.

Tool description rewritten to forbid preemptive calls and to make
`changed: false` after a failure the stop signal.

Tests: scripted stdio harness (run_stdio_credential_script) driving the
stackql_auth_testing dummy-credential provider; robot scenarios for
fresh-session scoped/unscoped/unknown, rotation after a registered
context (stale key rejected by the mock, new key used post-reload),
rotation before first query, recovery from failed resolution (select
and mutation error text, no-op reload), missing env file. The
pull_provider scenario now runs against a mocked HTTP registry server
(9927) so the install is real rather than a swallowed local-mode error.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

* mcp: master tool descriptions as markdown under content/tools

Tool descriptions were inline Go string literals spread across server.go
and query_library.go while instructions, prompts and resources were
already authored as markdown under pkg/mcp_server/content. Move the 16
descriptions to content/tools/<tool_name>.md (frontmatter `name`, body
is the description prose; soft line breaks collapse to spaces, blank
lines keep paragraph breaks) and load them through the same embedded
content framework.

addToolWithGate attaches the mastered description and returns an error
when a registered tool has none, so registration failures surface at
server construction like prompt and resource failures do. Published
bytes are unchanged for every tool. The build-time gate test now checks
coverage in both directions: every published tool has a file and every
file names a published tool.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

* mcp: test that markdown tool description prose is what tools/list publishes

Reads content/tools/run_select_query.md from disk independently of the
loader, strips the frontmatter and joins the wrapped lines, then compares
with the description returned by an in-process tools/list.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

* robot: assert server/discover serves instructions on the stateless HTTP server

The 2026-07-28 stateless roundtrip now opens with server/discover, the
SEP-2575 replacement for the initialize handshake, and the 9926
scenario asserts the response lists the revision and carries the
embedded content/instructions text. Closes the black-box gap where the
new-revision instructions path was only covered by the in-process unit
test.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

* tool descriptions

* mcp: align server instructions with the on-demand tool descriptions

server_info and reload_credentials are on-demand recovery and
diagnostic tools, never session-start steps; the instructions said
otherwise. Session guidance now states the defaults (sqlite3, safe),
that credentials resolve at query time with no preceding step, and the
fail -> fix file -> reload -> retry sequence with changed: false as the
stop signal. The 401 recovery rule gates the retry on changed: true.
Discovery points at the query library before the drill-down, dialect
rules describe validate_select_query as an on-demand pre-flight rather
than a routine step, and the mutation shapes gain the EXEC statement
form behind run_lifecycle_operation.

validate_select_query: the "no credentials required" claim was checked
against the binary and is false (EXPLAIN resolves credentials and fails
before reporting a bad column when they are missing); the description
now says so. Trailing newlines restored on the tool files.

The stateless discover scenario anchors on the "# Discovery workflow"
heading rather than a sentence that the rewrite reworded.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

* test: retire the bespoke credential reload roundtrip in favour of the scripted driver

The Sources Env File Mid Session scenario now runs through
run_stdio_credential_script like the other reload scenarios, and the
superseded run_stdio_credential_reload_roundtrip helper is removed. This
also clears the CodeQL clear-text-storage alert (pre-existing on main)
that the helper's secret_value write carried.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

---------

Co-authored-by: Claude Fable 5.1 <[email protected]>