/accept — OKS Modular Acceptance

Validate feasibility and component boundaries before proposing product changes. Never modify an existing knowledge base, its scheduler, or OpenClaw.

Run a non-Feishu capability

Run the helper on Linux/macOS from a controlled acceptance root. Supply the exact wheel or Git package reference being tested; do not silently test a different installed version.

python .claude/skills/accept/scripts/accept.py document \
  --root /opt/oks-acceptance \
  --package-spec /path/to/open_knowledge_studio-*.whl

When a server cannot fetch the fixed public fixture, copy an approved local sample into the isolated server area and provide it explicitly. This records a local_fixture source kind and SHA-256; it does not turn a network failure into a product pass.

python .claude/skills/accept/scripts/accept.py all ... \
  --fixture document=/opt/oks-acceptance/fixtures/sample.pptx \
  --fixture pdf=/opt/oks-acceptance/fixtures/digital.pdf \
  --fixture formula=/opt/oks-acceptance/fixtures/scanned-formula.pdf \
  --fixture watch=/opt/oks-acceptance/fixtures/short-video.mp4

Supported capabilities are document, pdf, formula, watch, feishu, and all. all is strictly serial: document → pdf → formula → watch → Feishu preflight. The final Feishu E2E remains awaiting_human until its dedicated Base and reviewer approval are supplied.

The helper creates one isolated pipx environment and one empty KB per ingest capability. Formula acceptance verifies capability installation and paddleocr import only — it is a secondary PDF sub-capability with no standalone ingest route. Feishu preflight creates no cloud data. It verifies install, oks, capability installation, Raw Bundle artifacts, Candidate promotion, search, recall, and lint. It writes report.json and report.md, then removes only that run's pipx and KB unless --keep-environment is supplied. all additionally writes matrix.json and matrix.md with install isolation, execution, resource cost, cleanup, and observed blockers for every component.

Interpret results honestly

• passed: every asserted local contract completed.
• product_failure: a local command, artifact, or lifecycle assertion failed.
• environment_limited: DNS, download, rate-limit, platform anti-bot, timeout, or unavailable external model prevented completion.
• awaiting_human: a required review or credential-gated step remains.

The public fixture URLs live in the helper. Reports record final source URLs, SHA-256 values, command exit codes, elapsed time, and cleanup outcome. They must never include environment-variable values, tokens, passwords, or private keys.

Run Feishu E2E

Feishu E2E creates a new dedicated Base and writes real test records. Before running it, obtain explicit authorization for that creation and export only runtime credentials:

export OKS_FEISHU_BASE_TOKEN='...'
export OKS_FEISHU_TABLE_ID='...'

Use a new Base created by python examples/feishu-loop/code/feishu_setup.py, submit the public text fixture, run one bounded worker pass, generate a Candidate from the Raw Bundle, and use bounded python examples/feishu-loop/code/feishu_base_worker.py listen-reviews while the designated reviewer sends an approval. Verify promotion and recall. Record only redacted identifiers. Do not delete the new Base automatically: report its identifier and wait for human confirmation.

After every run

Read the report before changing product code. Convert only observed product_failure entries into a separate proposal containing the command, artifact path, root cause, smallest fix, and regression test.