Skip to content

ci(core-api): freeze v1 broker OpenAPI baseline + oasdiff gate (Gate C Phase 1) - #620

Merged
1 commit merged into
mainfrom
CAURA-000-coreapi-openapi-oasdiff-gate
Jul 26, 2026
Merged

1 commit merged into
mainfrom
CAURA-000-coreapi-openapi-oasdiff-gate

Conversation

@ghost

@ghost ghost commented Jul 24, 2026

Copy link
Copy Markdown

What

Freeze the v1 broker↔cloud OpenAPI contract for core-api and enforce it on every PR with an oasdiff breaking-change gate. Phase 1 (Deliverables A+B) of the broker API-versioning RFC (docs/design/broker-cloud-api-versioning.md, enterprise repo).

Why

v1 was a label, not an enforced contract — breaking changes shipped inside /v1 and silently broke installed brokers (the Phase-0 incident). This commits the broker-facing contract as an artifact and fails CI on any breaking change to it, so a break must ship a new /vN.

Contents

  • core-api/scripts/gen_broker_openapi.py — imports core_api.app offline, filters paths to the 4 broker endpoints (POST /api/v1/memories/bulk, POST /api/v1/search, GET /api/v1/health, GET /api/v1/version; a missing path is a hard error), keeps only the transitively-reachable schemas + referenced security schemes, normalizes info to a frozen contract version (v1) decoupled from the rolling package semver (so releases don't churn the baseline), and emits deterministic JSON. --check fails if the committed baseline is stale.
  • core-api/openapi.broker.json — the frozen v1 baseline (4 paths, 14 schemas, APIKeyHeader).
  • ci.yml — stale-baseline guard + oasdiff breaking --fail-on ERR vs main's baseline (pinned tufin/oasdiff:v1.26.0; --fail-on ERR required — plain oasdiff breaking exits 0 even on breaks). First-merge bootstrap skips the diff when main has no baseline yet.
  • core-api/scripts/README.md — regeneration + contract-bump rules.

Validation

--check clean and deterministic (byte-identical across runs); proof the gate bites: a synthetic required field on BulkMemoryCreate → oasdiff reports new-required-request-property, exit 1 — then reverted. ruff clean on the script.

Regenerate after an intentional broker contract change: python core-api/scripts/gen_broker_openapi.py.

Gate C / Phase 1. Companion to the enterprise auth/audit gate (#889) and the Phase-0 fixes.

@github-actions

Copy link
Copy Markdown
Contributor

Claude Code Review — skipped: PR author 'Eldad-Caura' is not a public member of the 'caura-ai' org

@ghost

ghost commented Jul 24, 2026

Copy link
Copy Markdown
Author

@claude

@github-actions

Copy link
Copy Markdown
Contributor

Summary

This PR adds an API-versioning gate (Gate C / Phase 1) that enforces a frozen v1 broker OpenAPI contract via a freshness check and a breaking-change check using oasdiff. The script logic for transitive ref traversal and security scheme extraction is correct. Two real issues warrant attention before merge.


Medium Issues

Docker image pinned by mutable tag

Severity: Medium
File: .github/workflows/ci.yml:107
Problem: tufin/oasdiff:v1.26.0 is a mutable tag — it can be silently overwritten, and for a public repo where CI runs on fork PRs, a compromised tag executes arbitrary code in the runner.

🤖 Claude Code Prompt
In .github/workflows/ci.yml, the oasdiff Docker image is pinned by mutable tag:

  tufin/oasdiff:v1.26.0

Pin it by immutable digest instead. Fetch the digest for tufin/oasdiff:v1.26.0
(e.g. `docker inspect --format='{{index .RepoDigests 0}}' tufin/oasdiff:v1.26.0`
or via Docker Hub) and change the run step to:

  tufin/oasdiff:v1.26.0@sha256:<digest>

This prevents a tag overwrite from executing arbitrary code in CI for fork PRs.

Low Issues

Denylist for top-level OpenAPI keys silently pollutes the frozen contract

Severity: Low
File: core-api/scripts/gen_broker_openapi.py:160-170
Problem: spec = {k: v for k, v in full.items() if k not in ("paths", "components")} copies every unknown top-level key from the live app spec (e.g., tags, servers, webhooks) into the frozen contract — if FastAPI ever emits such keys, the broker baseline silently picks them up without a deliberate decision, causing unexpected --check failures on future PRs.

🤖 Claude Code Prompt
In core-api/scripts/gen_broker_openapi.py, build_broker_spec() uses a denylist
to forward top-level OpenAPI keys:

  spec = {k: v for k, v in full.items() if k not in ("paths", "components")}

Replace with an explicit allowlist of the only top-level keys that belong in the
frozen broker contract:

  ALLOWED_TOP_LEVEL = {"openapi", "security"}
  spec = {k: v for k, v in full.items() if k in ALLOWED_TOP_LEVEL}
  spec["info"] = {"title": CONTRACT_TITLE, "version": CONTRACT_VERSION}
  spec["paths"] = paths
  if components:
      spec["components"] = components

This ensures future additions to the full app spec (tags, servers, webhooks, etc.)
can never silently appear in the frozen contract without a deliberate code change here.

🟠 High Fix All

(no high issues)

🟡 Medium/Low Fix All

Fix 1 — core-api/scripts/gen_broker_openapi.py, build_broker_spec() (~line 160):
Replace:
  spec = {k: v for k, v in full.items() if k not in ("paths", "components")}
With:
  ALLOWED_TOP_LEVEL = {"openapi", "security"}
  spec = {k: v for k, v in full.items() if k in ALLOWED_TOP_LEVEL}
  spec["info"] = {"title": CONTRACT_TITLE, "version": CONTRACT_VERSION}
  spec["paths"] = paths
  if components:
      spec["components"] = components
(Remove the separate spec["info"]/spec["paths"]/spec["components"] assignments
that follow, since they are now inline above.)

Fix 2 — .github/workflows/ci.yml (~line 107):
Pin the oasdiff Docker image by immutable digest:
  Run: docker pull tufin/oasdiff:v1.26.0 && docker inspect --format='{{index .RepoDigests 0}}' tufin/oasdiff:v1.26.0
  Then replace `tufin/oasdiff:v1.26.0` with `tufin/oasdiff:v1.26.0@sha256:<digest>`

Reviewed by claude-sonnet-4-6 · cost $0.18834754999999997

@ghost
ghost force-pushed the CAURA-000-coreapi-openapi-oasdiff-gate branch from 07a0b9f to eb83835 Compare July 24, 2026 21:33
@github-actions

Copy link
Copy Markdown
Contributor

Claude Code Review — skipped: PR author 'Eldad-Caura' is not a public member of the 'caura-ai' org

@ghost

ghost commented Jul 24, 2026

Copy link
Copy Markdown
Author

@claude

@github-actions

Copy link
Copy Markdown
Contributor

Summary

This PR adds a CI gate for broker↔cloud API versioning: a generation script, a committed baseline (openapi.broker.json), and two CI steps (freshness check + oasdiff breaking-change check). The CI plumbing is well-constructed and the oasdiff digest-pin is a nice security touch. One real contract correctness issue exists: the frozen broker schema allows memory types the server will actively reject.


Critical/High Issues

(none)


Medium/Low Issues

BulkMemoryItem.memory_type schema allows server-rejected types

Severity: Medium
File: core-api/openapi.broker.json:249-265 / core-api/src/core_api/schemas.py (BulkMemoryItem)
Problem: The broker contract declares memory_type as $ref: MemoryType, whose enum carries all 14 values including outcome, rule, and insight — which the server rejects with HTTP 422 at the bulk-write boundary (memories.py:943 calls _reject_reserved_memory_type), so brokers that send any of those three types will always fail despite passing schema validation.

The field description correctly lists only the 10 writable types (it uses MEMORY_TYPES_WRITE_DESCRIPTION), but the schema reference to the full MemoryType enum contradicts it. Any JSON-schema-validating broker will treat all 14 as legal inputs.

semantic is also in the full enum but is silently remapped rather than rejected, so it's not a 422 risk — only outcome, rule, insight are actively refused.

🤖 Claude Code Prompt
In core-api/src/core_api/schemas.py, find the BulkMemoryItem model and its `memory_type` field. It is currently typed as `MemoryType | None`, where MemoryType is the full 14-value enum. The server rejects `outcome`, `rule`, and `insight` with HTTP 422 at the API boundary (see core-api/src/core_api/routes/memories.py _reject_reserved_memory_type called at line 943 for bulk items), so these types must not appear in the broker contract's enum.

Fix: in core-api/src/core_api/schemas.py, change the `memory_type` field on BulkMemoryItem to use a Literal of only the writable types (defined by MEMORY_TYPES_WRITE in core-api/src/core_api/constants.py — that tuple already excludes SERVER_RESERVED_MEMORY_TYPES and CLASSIFIER_DEPRECATED_MEMORY_TYPES). Keep the field description as-is (it already uses MEMORY_TYPES_WRITE_DESCRIPTION).

After fixing the schema, regenerate the baseline:
  PYTHONPATH=core-api/src:core-storage-api/src:. uv run python core-api/scripts/gen_broker_openapi.py

Commit both core-api/src/core_api/schemas.py and the updated core-api/openapi.broker.json.

--out silently ignored when --check is also passed

Severity: Low
File: core-api/scripts/gen_broker_openapi.py:162-180
Problem: If a caller passes both --check and --out PATH, --out is silently discarded; the caller gets no indication the flag was ignored.

🤖 Claude Code Prompt
In core-api/scripts/gen_broker_openapi.py in the main() function, after parsing args, add a guard before the `if args.check:` block:

  if args.check and args.out:
      print("--out is ignored when --check is given", file=sys.stderr)
      return 1

This makes the conflict explicit rather than silently dropping --out.

🚀 Fix All Prompts

🟡 Medium/Low

1. (Medium) In core-api/src/core_api/schemas.py, find BulkMemoryItem and change the `memory_type` field
   type from `MemoryType | None` to a Literal built from MEMORY_TYPES_WRITE (imported from
   core_api.constants). MEMORY_TYPES_WRITE already excludes SERVER_RESERVED_MEMORY_TYPES
   (outcome, rule, insight) and CLASSIFIER_DEPRECATED_MEMORY_TYPES (semantic). Keep the
   field description pointing at MEMORY_TYPES_WRITE_DESCRIPTION unchanged.
   After the schema change, regenerate and commit the baseline:
     PYTHONPATH=core-api/src:core-storage-api/src:. uv run python core-api/scripts/gen_broker_openapi.py

2. (Low) In core-api/scripts/gen_broker_openapi.py main(), add before the `if args.check:` block:
     if args.check and args.out:
         print("--out is ignored when --check is given", file=sys.stderr)
         return 1

Reviewed by claude-sonnet-4-6 · cost $0.4391463

@ghost
ghost force-pushed the CAURA-000-coreapi-openapi-oasdiff-gate branch from eb83835 to 331cc3e Compare July 24, 2026 22:06
@github-actions

Copy link
Copy Markdown
Contributor

Claude Code Review — skipped: PR author 'Eldad-Caura' is not a public member of the 'caura-ai' org

@ghost
ghost force-pushed the CAURA-000-coreapi-openapi-oasdiff-gate branch from 331cc3e to 45d9ab0 Compare July 24, 2026 22:38
@github-actions

Copy link
Copy Markdown
Contributor

Claude Code Review — skipped: PR author 'Eldad-Caura' is not a public member of the 'caura-ai' org

@ghost

ghost commented Jul 24, 2026

Copy link
Copy Markdown
Author

Fixed the Low finding (--out + --check now errors explicitly instead of silently ignoring --out).

On the Medium (baseline advertises server-reserved memory_types outcome/rule/insight): the right fix is #621, which changes BulkMemoryItem.memory_type from the full MemoryType enum to str (additive-tolerance — unknown types become per-item 207s, matching content/weight/status). With that, this baseline regenerates to memory_type: {"type": "string"} — no reserved-type enum to mis-advertise. I deliberately did not put a conflicting Literal[MEMORY_TYPES_WRITE] change here.

Suggested order: merge #621 first (it's rebased onto main, clean, mergeable), then I'll rebase this PR and regenerate openapi.broker.json so the baseline reflects the str field — which resolves this Medium. @erni-a — flagging the sequencing; happy to do it the other way (Literal here, #621 supersedes) if you prefer.

erni-a
erni-a previously approved these changes Jul 24, 2026
@ghost
ghost force-pushed the CAURA-000-coreapi-openapi-oasdiff-gate branch from 45d9ab0 to b9e3490 Compare July 26, 2026 06:28
@github-actions

Copy link
Copy Markdown
Contributor

Claude Code Review — skipped: PR author 'Eldad-Caura' is not a public member of the 'caura-ai' org

@ghost
ghost force-pushed the CAURA-000-coreapi-openapi-oasdiff-gate branch from b9e3490 to 08609e5 Compare July 26, 2026 06:34
@github-actions

Copy link
Copy Markdown
Contributor

Claude Code Review — skipped: PR author 'Eldad-Caura' is not a public member of the 'caura-ai' org

@ghost
ghost force-pushed the CAURA-000-coreapi-openapi-oasdiff-gate branch from 08609e5 to 3606812 Compare July 26, 2026 06:39
@github-actions

Copy link
Copy Markdown
Contributor

Claude Code Review — skipped: PR author 'Eldad-Caura' is not a public member of the 'caura-ai' org

@ghost

ghost commented Jul 26, 2026

Copy link
Copy Markdown
Author

@claude rebased onto main after #621 merged and regenerated the baseline — memory_type in openapi.broker.json is now a plain string ($ref: MemoryType enum removed), so the contract no longer advertises the server-reserved types; that resolves the Medium. The Low finding (--out+--check) is fixed. gen --check passes locally (baseline current). Please re-verify.

@github-actions

Copy link
Copy Markdown
Contributor

The script doesn't exist on disk yet (PR not merged), so I'm working from the diff alone. I have enough to complete the review.


Summary

This PR adds a CI gate (freshness + breaking-change check via oasdiff) for a frozen v1 broker OpenAPI contract, a generator/verifier script, and the committed baseline. The CI structure is sound and the Docker image is correctly pinned by digest. One functional bug in the generator causes the broker contract to silently include security schemes it never actually uses.


Medium/Low Issues

Broker spec inherits unused global security schemes from the full spec

Severity: Medium
File: core-api/scripts/gen_broker_openapi.py:121-127 (diff lines ~121–127)
Problem: _security_scheme_names is called with full.get("security") — the full spec's global security block — which pulls schemes into the broker contract even if no broker endpoint references them, making the contract inaccurate and creating future false oasdiff breaking-change alerts.

# Current
scheme_names = _security_scheme_names(paths, full.get("security"))

The broker output spec never emits a top-level security field, so global security from the full spec is irrelevant. Any scheme reachable only via full["security"] (not via an individual broker operation's security array) gets added to securitySchemes, and if it's later removed from the full spec, oasdiff will flag it as a breaking broker-contract change even though no broker endpoint ever used it.

🤖 Claude Code Prompt
In core-api/scripts/gen_broker_openapi.py, the `build_broker_spec` function calls:

    scheme_names = _security_scheme_names(paths, full.get("security"))

The second argument passes the *full* OpenAPI spec's top-level `security` block. The broker subset spec does not emit a top-level `security` field, so schemes referenced only globally (not by any broker operation's explicit `security` array) get included in `securitySchemes` without being referenced by any broker path. This means:
1. The broker contract becomes inaccurate (it contains schemes no broker operation uses).
2. If those schemes are later removed from the full spec, oasdiff incorrectly flags the removal as a breaking broker-contract change.

Fix: pass `None` instead of `full.get("security")` so only operation-level security from the broker paths contributes to `scheme_names`:

    scheme_names = _security_scheme_names(paths, None)

The `_security_scheme_names` function already iterates each operation's `security` array from the `paths` argument, which covers all broker-endpoint security requirements.

🟡 Medium/Low

In core-api/scripts/gen_broker_openapi.py, fix the security-scheme collection so it only
draws from broker operation-level `security` blocks, not the full spec's global security:

Change:
    scheme_names = _security_scheme_names(paths, full.get("security"))

To:
    scheme_names = _security_scheme_names(paths, None)

Rationale: the broker spec does not emit a top-level `security` field, so the full spec's
global security block is irrelevant. Passing it caused schemes unreferenced by any broker
operation to appear in `securitySchemes`, which would cause oasdiff false positives if
those schemes are later removed from the full spec.

Reviewed by claude-sonnet-4-6 · cost $0.2533269

@ghost

ghost commented Jul 26, 2026

Copy link
Copy Markdown
Author

@claude fixed: _security_scheme_names now collects each broker operation's effective security (its own security, else the spec default) instead of unconditionally unioning the global security block — so a scheme in the global block that every broker op overrides no longer leaks into the contract, and there's no risk of a dangling top-level security ref. Current spec is unaffected (global security is None; bulk/search declare APIKeyHeader, health/version are unauthenticated), so the regenerated baseline is byte-identical. gen --check + ruff clean.

@ghost
ghost force-pushed the CAURA-000-coreapi-openapi-oasdiff-gate branch from 3606812 to 927228a Compare July 26, 2026 06:50
@github-actions

Copy link
Copy Markdown
Contributor

Claude Code Review — skipped: PR author 'Eldad-Caura' is not a public member of the 'caura-ai' org

@github-actions

Copy link
Copy Markdown
Contributor

The script is new in this PR so I'll review from the diff alone.


Summary

Adds a two-step CI gate that freezes the four broker-facing gateway endpoints as core-api/openapi.broker.json: a freshness check (gen_broker_openapi.py --check) and a breaking-change check (oasdiff breaking against main). The transitive-closure ref-walking logic is correct and the serialisation is deterministic. One structural fragility in the generator is worth fixing before this bakes in.

Medium/Low Issues

Top-level key passthrough uses a denylist instead of an allowlist

Severity: Low
File: core-api/scripts/gen_broker_openapi.py:143
Problem: All top-level keys from app.openapi() except paths and components are passed through verbatim, so any key the FastAPI app gains in the future (servers, tags, externalDocs) will silently appear in the frozen broker baseline, causing spurious oasdiff noise or stale-baseline CI failures with no broker-endpoint change.

🤖 Claude Code Prompt
In core-api/scripts/gen_broker_openapi.py, build_broker_spec() builds the
broker spec with a denylist:

    spec = {k: v for k, v in full.items() if k not in ("paths", "components")}

This lets any future top-level key added to app.openapi() (e.g. `servers`,
`tags`, `externalDocs`) flow silently into the frozen broker baseline.

Replace that one line with an explicit allowlist:

    spec = {"openapi": full["openapi"]}

The subsequent lines that set spec["info"], spec["paths"], and spec["components"]
are unchanged and produce identical output for the current app (which only
emits `openapi`, `info`, `paths`, `components`).  If a new top-level key is
ever needed in the broker contract it must be added consciously here rather
than leaking in automatically.

After the change, regenerate and verify the baseline is unchanged:
    python core-api/scripts/gen_broker_openapi.py --check

🟢 Fix All (Low)

In core-api/scripts/gen_broker_openapi.py ~line 143, replace the denylist:

    spec = {k: v for k, v in full.items() if k not in ("paths", "components")}

with an explicit allowlist:

    spec = {"openapi": full["openapi"]}

The three lines that follow (setting spec["info"], spec["paths"],
spec["components"]) are already correct and unchanged.

Verify the baseline is still current after the edit:
    PYTHONPATH=core-api/src:core-storage-api/src:. \
        python core-api/scripts/gen_broker_openapi.py --check

Reviewed by claude-sonnet-4-6 · cost $0.39624889999999996

…ange gate

Phase 1 of the broker<->cloud API-versioning RFC (freeze + enforce v1):
commit the immutable v1 OpenAPI contract for core-api's four broker-facing
gateway endpoints (POST /api/v1/memories/bulk, POST /api/v1/search,
GET /api/v1/health, GET /api/v1/version) that memclawd calls against cloud,
and gate PRs on it.

- core-api/scripts/gen_broker_openapi.py: imports core_api.app offline, filters
  paths to the 4 broker endpoints (missing path = hard error), keeps only the
  transitively-reachable schemas + referenced security schemes, normalizes info
  to a frozen contract identity (version "v1", decoupled from the rolling
  package semver so releases don't churn it), and emits deterministic JSON.
  --check fails if the committed baseline is stale vs the code.
- core-api/openapi.broker.json: the frozen v1 baseline (4 paths, 14 schemas).
- ci.yml: a stale-baseline guard + `oasdiff breaking --fail-on ERR` vs main's
  baseline (pinned tufin/oasdiff:v1.26.0). --fail-on ERR is required — plain
  `oasdiff breaking` exits 0 even on breaks. First-merge bootstrap skips the
  diff when main has no baseline yet.
- core-api/scripts/README.md documents regeneration + the contract-bump rule.

A breaking change to a shipped v1 broker endpoint now fails CI; per the RFC it
must ship a new /vN version rather than edit the frozen baseline.

Gate C / Phase 1 (Deliverables A+B).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Signed-off-by: eldad-caura <eldad@caura.ai>
@ghost
ghost force-pushed the CAURA-000-coreapi-openapi-oasdiff-gate branch from 927228a to 38c0fc3 Compare July 26, 2026 07:09
@github-actions

Copy link
Copy Markdown
Contributor

Claude Code Review — skipped: PR author 'Eldad-Caura' is not a public member of the 'caura-ai' org

@ghost

ghost commented Jul 26, 2026

Copy link
Copy Markdown
Author

@claude fixed the Low: build_broker_spec now uses an explicit top-level allowlist ({"openapi": ...} + info/paths/components) instead of the denylist, so a future top-level key from app.openapi() can't leak into the frozen baseline. Regenerated — baseline byte-identical; gen --check + ruff clean.

@ghost

ghost commented Jul 26, 2026

Copy link
Copy Markdown
Author

@erni-a please approve

@ghost
ghost merged commit 1d08473 into main Jul 26, 2026
10 checks passed
@ghost
ghost deleted the CAURA-000-coreapi-openapi-oasdiff-gate branch July 26, 2026 07:13
ran-taig added a commit that referenced this pull request Jul 26, 2026
…625)

## Summary

Executes all remaining OSS-side work from the 2026-07-26
product-coordination audit (enterprise-side fixes landed in the
enterprise repo's #914). One commit per task, all DCO-signed.

- **Canonical SKILL.md** (`static/skills/memclaw/SKILL.md`): documents
the Interviewer (§3 — reflective server-side synthesis; realtime writes
stay mandatory) and the Skill Factory (§8 — `staged` is governance not
an error, `forge/`/`agent/` `doc_id` provenance, push delivery). Five
new content-invariant pins in `tests/test_direct_mcp_skill.py`. Once
merged, the enterprise nightly `sync-tools-json.yml` carries these
sections to `memclaw.net/docs/agents.md` via its pin-bump PR — no manual
step.
- **core-operations docstring**: `app.py` no longer claims "this
scaffold registers nothing" — it names the nine registered cron jobs
(six daily lifecycle ticks, daily/weekly agent digests, hourly
interviewer-schedule), each verified against its `scheduler.register`
call.
- **README**: new **MemClaw Broker** section (identity headers,
`broker:<install>` namespace, frozen v1 wire contract + oasdiff gate
from #620); agent-digests bullet (`agent_digest.enabled`, `GET
/api/v1/reports*`); a deployment-topology table stating exactly what
`docker compose up` starts (db, redis, core-storage-api, core-api; `tei`
only with `--profile embed-local`) vs what is managed-only (core-worker,
platform tier, Pub/Sub); clarifies the `core-worker` backfill variant
isn't in the stock compose stack.
- **Plugin SKILL.md variant**: ports both notes, with the Interviewer
explicitly distinguished from the plugin's `MEMCLAW_AUTO_WRITE_TURNS`
auto-write layer (both are floors, not substitutes); adds a mirroring
content-invariant test to `plugin/src/tool-definitions.test.ts`.
- **Component registry** (P2): `docs/component-registry.yaml` (4
services, 9 cron ticks, 5 event-topic families, Forge external-scheduler
requirement) + `tests/test_component_registry.py`, an AST-based two-way
drift gate between `scheduler.register(...)` calls and the registry —
F5-class drift now fails a test.

## Test plan

- `python3 -m pytest tests/test_direct_mcp_skill.py -q` → 10 passed
- `python3 -m pytest tests/test_component_registry.py -q` → 3 passed
- `cd plugin && npm test` → 378 passed, 0 failed
- Full root suite: 3964 passed; 5 failures are local-env only (missing
`google-cloud-pubsub`, stray `.claude/worktrees` copies) — none touch
this diff
- `ruff check` / `ruff format --check` clean on the touched Python
files; no `core-api`/`core-storage-api` src changes, so CI ruff/mypy
surfaces are untouched

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Signed-off-by: Ran Taig <ran@caura.ai>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
This pull request was closed.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants