Digitize outsourcing (Odoo → outside digitizer)

You run the embroidery digitizing outsourcing loop. Embroidery artwork must be digitized (turned into a stitch file) before it can be sewn; when that work goes to an outside digitizer, the record that tracks it is an Odoo digitize.demand. Your job, on a routine, is to send the Ready ones out and mark them Outsourced.

This is the ERP half of the loop. The portal half — actually posting the job into the digitizer's web form — is the separate artworklady-digitizing skill, which you invoke. Keep the two responsibilities separate: this skill decides what to send and updates Odoo; that skill sends it.

SEND leg:
digitize_demand_search (Ready, non-custom)  →  pull decoration art + specs
      →  download the PNG + AI files to disk
      →  build the brief (H × W + thread codes + colours)
      →  artworklady-digitizing submit (dry run, then confirm; standard turnaround)
      →  on status=="submitted":  digitize_demand_outsource   (Ready → Outsource)

RETURN leg:
digitize_demand_search (state=outsource)  →  fetch the vendor's returned files
      →  file them on your machine         (persistent job archive + manifest)
      →  digitize_demand_upload            (Outsource → Sew Out)
      →  parse_worksheet (the PDF)         →  stitch count, size, Stop-Sequence colours
      →  digitize_demand_apply_worksheet   (dims + stitch count + thread colours)

Transport

Odoo is driven through the drivethru_mcp MCP server via scripts/odoo_mcp.py:

python3 scripts/odoo_mcp.py tools                        # discover tools + schemas
python3 scripts/odoo_mcp.py call <tool> '{"...":"..."}'  # call a tool
echo '{"...":"..."}' | python3 scripts/odoo_mcp.py call <tool>

Every call prints one JSON object, or {"error": {...}} with a non-zero exit. ODOO_MCP_URL / ODOO_MCP_TOKEN must be set — if not, stop and tell the user. Run tools first and trust its inputSchema over any signature below.

The routine, step by step

1. Read the Ready queue

python3 scripts/odoo_mcp.py call digitize_demand_search '{"is_custom": false}'

This returns Ready, non-custom demands — the queue (custom-text demands use a different digitizing flow, so exclude them). Send only demands whose decoration_method is exactly "Embroidery" — skip "Embroidery (Tackle Twill)", Puff/3D embroidery, and any other variant. Also skip any demand whose decoration already has a Shopworks ID (shopworks_design_number set): those are already digitized in Shopworks and only need their file uploaded, not outsourcing — report them so a human can upload, but never send them. Read the roll-up (sendable_count, blocked_count, rush_count) and each row. Work at most ~20 per run, soonest earliest_ship_date first; anything beyond that is picked up next run:

Skip anything already outsourced. The queue is Ready-only by default, so a demand that has been sent is already gone from it; never widen with include_all/is_outsourced and then send those.

2. Download the two files

For each sendable demand, get the decoration's embroidery production PNG and its AI source file with digitize_demand_files — one call returns both:

python3 scripts/odoo_mcp.py call digitize_demand_files '{"demand_id": <DEMAND_ID>}'

Each entry comes back with filename, mimetype, size_bytes, source, and data_base64 (or error). Decode each data_base64 to a file (e.g. emb_<demand_id>.png, emb_<demand_id>.ai) and confirm it's non-empty.

Do not try to fetch the row's art.*.cdn_url yourself: on production those binaries live on a CDN firewalled to Odoo's IP, so an agent can't reach it — that's exactly why digitize_demand_files fetches them server-side and hands you the bytes. If a file comes back with error (missing, or over the 20 MB guard), that demand isn't sendable — report it and move on. See references/routine.md for details.

3. Rush is informational — submit everything, standard turnaround

is_rush (ship date within the window) and days_until_ship are surfaced per row, but the routine does not act on them right now. It submits every outsource_ready, non-custom demand, and always at the standard nextday turnaround — it does not select Artwork Lady's rush option. (The current backlog trips is_rush on old ship dates that aren't true rushes, so a rush turnaround would be wrong and can cost more.) Report the rush rows for visibility, but don't treat them differently. Revisit if genuine rush handling is wanted later.

4. Build the brief

The message you send the digitizer must clearly state the height × width and the thread colours — each with its Odoo colour code. The code is essential: a bare name like "White" lets the digitizer substitute a different thread code than the one the shop picked (and the return leg then records their substitution over your choice). Build the colour list from art.colors (each {name, color_code, pms_color}), not color_names — format each as <color_code> <name>:

Embroidery digitizing. Size: 3.5in W × 2.0in H.
Thread colours — use these exact codes: 1801 White, 1765 Tennessee Orange.
Left chest. Digitize to DST.

Build the size from art.width_in / art.height_in. If a colour has no color_code on the row, use the name alone and flag it (the digitizer will pick a code, which the return leg then records). Add placement/decoration context if the row carries it.

5. Submit to the digitizer (dry run, then confirm)

Hand the actual submission to the artworklady-digitizing skill. Its creds (ARTWORKLADY_USERNAME/PASSWORD, and, behind the egress proxy, the TLS SPKI pins) are that skill's concern — if it exits config_error, stop and tell the user to configure them; never ask for them in chat.

Always dry-run first (default confirm:false) to validate the fill, then submit with confirm:true:

python3 <artworklady-digitizing>/scripts/artworklady_digitizing.py submit '{
  "job_reference": "<DECO_ID>",
  "turnaround": "nextday",
  "file_types": ["dst"],
  "files": ["/abs/emb_<id>.png", "/abs/emb_<id>.ai"],
  "message": "Embroidery digitizing. Size: 3.5in W × 2.0in H. Thread colours — use these exact codes: 1801 White, 1765 Tennessee Orange. …"
}'

• job_reference = the decoration id (what that skill expects).
• files — the PNG first (the mandatory upload slot), the AI second. Both go up.
• turnaround — always nextday for now; do not select rushorder / sameday (see §3). Revisit only if genuine rush handling is turned on.
• Success is judged on the real submit response (HTTP 200 → status:"submitted"). Treat only status:"submitted" as sent — never a dry_run, an error, or the sight of /thanks.

6. Mark it Outsourced — exactly once

Only on status:"submitted", transition the demand:

python3 scripts/odoo_mcp.py call digitize_demand_outsource \
  '{"demand_id": <DEMAND_ID>, "note": "Artwork Lady — job_reference <DECO_ID>, submitted <ISO time>"}'

This moves the demand ready → outsource, stamps outsource_date, and flags is_outsourced on the demand and the decoration. Do this immediately after each confirmed submission, one demand at a time.

digitize_demand_outsource is idempotent (a demand already outsourced comes back as already_outsourced), but the window between a confirmed send and this call is the one risk: the portal returns no id, so a resend can't be detected. If this call fails after a confirmed submit, stop and report it — do not re-submit the job.

Idempotency & exactly-once (read this)

• Work only state=ready demands. Once sent they're at outsource and drop out of the default queue, so a re-run never re-sends them.
• Submit first, mark outsourced after a submitted result — and put the outsource call right after the submit for each demand, so a crash leaves the demand Ready (correct: it was never actually sent) rather than sent-but-unmarked.
• The portal does not dedupe and returns no id. The demand's state + the chatter note (job reference + submitted-at) are your audit trail.

Return leg: receive the digitized file (Outsource → Sew Out)

When the digitizer finishes, pull the stitch file back and complete the demand.

1. Find the demands waiting on the digitizer:

python3 scripts/odoo_mcp.py call digitize_demand_search '{"state":["outsource"]}'

Process only demands whose decoration_method is exactly "Embroidery" (skip Tackle Twill / Puff / other variants). Work at most ~20 per run; the rest come next hour.

2. Fetch the vendor's returned files from the portal with the artworklady-digitizing skill's fetch command. It logs into our account, finds our order row on /orders by job reference (the decoration id), and downloads that row's own download links — so we get the files the vendor actually produced. Fetch straight into the job's archive folder so the files land where you keep them (see step 3):

python3 <artworklady-digitizing>/scripts/artworklady_digitizing.py fetch \
  '{"job_reference": "<DECO_ID>", "out_dir": "/home/agent/.openclaw/jobs/<DECO_ID>"}'

• status:"fetched" → take the stitch file from the dst path in the result, and the worksheet PDF from the .pdf entry in files (there's always one for the job) — you need it in steps 5–6.
• status:"pending" → the order isn't finished yet; skip and retry next pass.
• status:"not_found" → no order for that job reference; report it.

Do not guess the file URL. The vendor's file store is shared across customers, so a job-number collision can rename the file (e.g. 8524_2.DST); scraping our own order row is what guarantees the right file.

3. File the vendor's return package on your own machine. Keep everything the order returned (typically the worksheet PDF, the DST, the native .EMB design, a .TBF) in a per-job folder on persistent storage, and drop a manifest.json beside it so the folder is self-describing:

# the archive folder IS the fetch out_dir above — persistent, survives redeploys
JOB=/home/agent/.openclaw/jobs/<DECO_ID>
cat > "$JOB/manifest.json" <<JSON
{"decoration_id": <DECO_ID>, "job_reference": "<DECO_ID>",
 "customer": "<customer>", "job_name": "<decoration name>",
 "digitize_demand_id": <DEMAND_ID>, "artwork_lady_order_id": "<order # if known>",
 "date_filed_utc": "<UTC now>", "files": [{"filename": "...", "bytes": 0}]}
JSON

• Persistent, not /tmp. The archive must live on durable storage — for this agent that's the mounted volume under /home/agent/.openclaw/…; anything under /tmp is wiped on redeploy. If your persistent workspace is elsewhere, use that (and use .../vendor-files/<DECO_ID>/ if jobs/ is taken).
• This is deliberately not stored in Odoo — it's the shop's own copy of what the vendor sent, like a job folder on an employee's work machine. When someone asks for a job's files later, look here by decoration id.
• Re-filing a job just overwrites its folder, so a re-run is safe.

4. Upload the DST (and the worksheet PDF) back to Odoo, which writes them onto the decoration and advances the demand outsource → sew_out:

DST_B64=$(base64 -w0 "$JOB"/*.DST)
PDF_B64=$(base64 -w0 "$JOB"/*.pdf)
python3 scripts/odoo_mcp.py call digitize_demand_upload \
  "{\"demand_id\": <DEMAND_ID>, \"dst_base64\": \"$DST_B64\", \"pdf_base64\": \"$PDF_B64\"}"

(Read them from the archive folder you filed in step 3, and pipe the JSON via stdin — the base64 is large.) digitize_demand_upload is idempotent — a demand already at Sew Out/Done comes back as already_uploaded, so a re-run skips finished work. Leave stitch_count to step 6, which is the single source of truth for the worksheet numbers.

5. Parse the worksheet PDF. scripts/parse_worksheet.py reads the stitch count, sewn size, and the Stop-Sequence thread list positionally (see references/routine.md):

python3 scripts/parse_worksheet.py "$JOB"/*.pdf --pretty
#   → {"stitch_count":5455,"height_in":2.08,"width_in":4.01,
#      "colors":[{"sequence":1,"code":"1765","name":"Tiger",…},
#                {"sequence":2,"code":"1801","name":"Super White",…}], "warnings":[]}

Read warnings — a colour count mismatch or a millimetre size means eyeball the PDF before applying. A row with no code can't be matched; surface it.

6. Apply the worksheet to the decoration. Pass the parsed values straight through — colors needs only code, name, sequence per row:

python3 scripts/odoo_mcp.py call digitize_demand_apply_worksheet '{
  "demand_id": <DEMAND_ID>,
  "stitch_count": 5455, "height_in": 2.08, "width_in": 4.01,
  "colors": [{"sequence":1,"code":"1765","name":"Tiger"},
             {"sequence":2,"code":"1801","name":"Super White"}]
}'

This writes number_of_stitches, size_height/size_width, and rebuilds the decoration's thread colours in stitch order — matching each by its code. A code not on file is created (name + code) and a note is posted on the demand's batch @mentioning the thread buyer (default Laura McCarter) to buy the thread or change the colour. The result's colors.matched / colors.created and note tell you what happened; report any created colours so a human follows up. Re-running is safe — a created colour matches next time, so the note fires once. Dry-run first with "dry_run": true if you want to preview the matches and creates without writing.

A pass, end to end

Send leg:

1. digitize_demand_search '{"is_custom": false}' → the Ready, non-custom queue.
2. If sendable_count is 0, report blocked_count + top blockers and stop.
3. Keep only demands whose decoration_method is exactly "Embroidery" (skip Tackle Twill / Puff / other variants) and with no shopworks_design_number (already digitized in Shopworks — report for manual upload, don't send), then take up to ~20 outsource_ready, soonest earliest_ship_date first.
4. For each, one at a time: download PNG + AI → build the brief → artworklady-digitizing submit dry run → confirm (standard nextday turnaround) → on submitted, digitize_demand_outsource.
5. Report: how many outsourced, which were blocked (and why), and any rush rows (informational only — still sent at standard turnaround).

Return leg:

6. digitize_demand_search '{"state":["outsource"]}' → demands out at the digitizer. Process only Embroidery-method demands (skip Tackle Twill / Puff); cap ~20/run.
7. For each, artworklady-digitizing fetch into the job's archive folder (out_dir=/home/agent/.openclaw/jobs/<deco id>); skip any that come back pending (not finished yet).
8. File the vendor return package: the fetch already put the files in the archive folder — write a manifest.json beside them. Persistent storage, not /tmp; not Odoo.
9. base64 the archived .DST (+ the .pdf as pdf_base64) and digitize_demand_upload it → the demand moves to Sew Out.
10. parse_worksheet.py <archived .pdf> → digitize_demand_apply_worksheet with the stitch count, size, and Stop-Sequence colours. Report how many were received vs still pending, any thread colours that had to be created (batch note raised), and where each job was filed.

Errors

• config_error (exit 2) — ODOO_MCP_URL / ODOO_MCP_TOKEN missing (this skill), or ARTWORKLADY_USERNAME/PASSWORD missing (the digitizing skill).
• connection_error — Odoo unreachable or key rejected.
• A tool that ran but failed returns a normal MCP result with isError: true and a human-readable message — surface it.
• portal_error from the digitizing skill — the submit did not return HTTP 200; the demand stays Ready and is retried next pass. Do NOT mark it outsourced.

References

• references/routine.md — the download details, the turnaround/rush mapping, file naming, and the exactly-once reasoning in full.