Skip to content

docs: add AGENTS.md, agent skills config, ADRs, and upload helper - #20

Open
gdccyuen wants to merge 5 commits into
Ontos-AI:mainfrom
gdccyuen:docs/agent-skills-and-adrs
Open

docs: add AGENTS.md, agent skills config, ADRs, and upload helper#20
gdccyuen wants to merge 5 commits into
Ontos-AI:mainfrom
gdccyuen:docs/agent-skills-and-adrs

Conversation

@gdccyuen

Copy link
Copy Markdown

What

Adds agent infrastructure docs and an upload helper script to knowhere-self-hosted.

Changes

Agent skills config (AGENTS.md + docs/agents/)

  • AGENTS.md — compact instruction file for OpenCode sessions covering repo purpose, verification (smoke-test.sh), local image build flow, env layering gotchas (.env.defaults vs entrypoint.sh), single-container 3-process runtime, compose services, telemetry, and conventions. Also includes an ## Agent skills block pointing at the three config docs.
  • docs/agents/issue-tracker.md — local markdown tracker under .scratch/<feature-slug>/
  • docs/agents/triage-labels.md — canonical defaults (needs-triage, needs-info, ready-for-agent, ready-for-human, wontfix)
  • docs/agents/domain.md — single-context layout (CONTEXT.md + docs/adr/)
  • .gitignore — added .scratch/ so local planning files stay local

ADRs (docs/adr/)

  • 0001-local-mineru-mode.md — moved from repo root. Updated status to reference upstream PR Ontos-AI/knowhere#233. Added pitfalls 9–11 discovered during live testing against MinerU 3.4.0: files form field (not file), default JSON response (not ZIP), and response_format_zip=true behavior.
  • 0002-mineru-raw-zip-audit-archive.md — new ADR for permanently archiving raw MinerU ZIPs to S3 with the S3 key in job_results. Covers both local and cloud mode. 6-step implementation plan.

Upload helper (upload-pdf.sh)

Shell script that uploads a PDF via the API end-to-end: create job → upload to S3 → confirm upload. Usage: ./upload-pdf.sh <API_KEY> [namespace] [file_path].

gdccyuen added 5 commits July 29, 2026 17:50
… script

- AGENTS.md: compact instruction file for OpenCode sessions covering
  repo purpose, verification, build flow, env layering, runtime
  architecture, conventions, and agent skills config
- docs/agents/: issue tracker (local markdown under .scratch/), triage
  labels (canonical defaults), domain docs (single-context layout)
- docs/adr/0001-local-mineru-mode.md: ADR for local MinerU mode via
  forked image patch, moved from repo root
- .gitignore: add .scratch/ for local issue tracking
- upload-pdf.sh: helper script to upload a PDF via the API from start
  to confirm-upload
…w ZIP audit archive

ADR 0001 updated:
- Status: upstream PR #233 opened
- Pitfalls 9-11: files form field, JSON response (not ZIP),
  response_format_zip=true behavior
- Future work: updated to reflect current state

ADR 0002 added:
- Decision to permanently archive raw MinerU ZIPs to S3
- Covers both local and cloud mode
- Storage: S3 + DB column (not DB blob)
- 6-step implementation plan (todos 1/6-6/6)
- README (en/zh): add Input Format Recommendation — convert Office files
  to PDF before uploading. Documents why (docx pipeline flattens tables,
  drops footers, no full.md; MinerU office backend unreliable; empirical
  findings from funding-application testing) and the recommended workflow
  for cost/savings analysis
- upload-pdf.sh: fix usage to <API_KEY> [file_path] [namespace] with
  file_path required, namespace defaulting to 'default'
Live testing against real funding-application .docx files showed the
office backend returns empty markdown (only a page-number header
survives). EMF vector images are skipped by the image loader and the
docx converter emits zero paragraphs/tables even without EMFs.
Recommend converting Office files to PDF before uploading.
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.

1 participant