$npx -y skills add vaquarkhan/data-engineering-agent-skills --skill file-and-partner-feed-ingestionGuides agents through file-based and partner-feed ingestion workflows. Use when landing data from SFTP, managed file transfer, shared buckets, recurring flat files, manifests, or externally supplied feeds that need validation, replay safety, and publish discipline.
| 1 | # File And Partner Feed Ingestion |
| 2 | |
| 3 | ## Overview |
| 4 | |
| 5 | Use this skill when ingestion depends on externally delivered files rather than APIs or CDC streams. It helps agents design landing, validation, late-file handling, checksum or manifest controls, replay safety, and publish boundaries for partner-managed feeds. |
| 6 | |
| 7 | ## When to Use |
| 8 | |
| 9 | - onboarding `SFTP`, MFT, or shared-bucket file feeds |
| 10 | - ingesting recurring `CSV`, `JSON`, `XML`, or columnar extracts from external partners |
| 11 | - validating manifests, checksums, control totals, or arrival windows |
| 12 | - handling late, duplicate, partial, or corrected file deliveries |
| 13 | - defining replay-safe landing and publish behavior for externally supplied data |
| 14 | |
| 15 | Do not assume file ingestion is simple just because the payload arrives in a batch. |
| 16 | |
| 17 | ## Workflow |
| 18 | |
| 19 | 1. Define the feed contract. |
| 20 | Clarify: |
| 21 | - file format and expected schema |
| 22 | - naming convention |
| 23 | - arrival schedule or SLA |
| 24 | - manifest, checksum, or control-total expectations |
| 25 | - ownership and contact path for the partner or source team |
| 26 | |
| 27 | 2. Design the landing boundary. |
| 28 | Decide: |
| 29 | - where raw files land |
| 30 | - what metadata is captured |
| 31 | - how duplicates are detected |
| 32 | - how partial or corrupt deliveries are quarantined |
| 33 | |
| 34 | 3. Validate before publish. |
| 35 | Include: |
| 36 | - schema and required-field checks |
| 37 | - checksum or manifest validation |
| 38 | - row-count or control-total validation |
| 39 | - late-file and missing-file detection |
| 40 | |
| 41 | 4. Make replay behavior explicit. |
| 42 | Decide how: |
| 43 | - corrected files are handled |
| 44 | - duplicate files are ignored or reconciled |
| 45 | - replay windows are bounded |
| 46 | - downstream publishes stay protected until validation passes |
| 47 | |
| 48 | 5. Define failure and escalation behavior. |
| 49 | Document: |
| 50 | - who gets alerted |
| 51 | - what blocks publish |
| 52 | - how the partner is contacted |
| 53 | - how recovery evidence is recorded |
| 54 | |
| 55 | ## Common Rationalizations |
| 56 | |
| 57 | | Rationalization | Reality | |
| 58 | | --- | --- | |
| 59 | | "It is just a file drop." | File-based feeds often fail through late arrival, partial delivery, duplicate upload, or silent schema drift. | |
| 60 | | "We can validate after loading." | Weak landing controls let corrupt or incomplete partner data flow too far downstream. | |
| 61 | | "If the partner resends the file, we are fine." | Resends without replay-safe handling can create duplicate or conflicting publishes. | |
| 62 | |
| 63 | ## Red Flags |
| 64 | |
| 65 | - file naming and arrival expectations are undocumented |
| 66 | - manifest, checksum, or control totals are ignored |
| 67 | - duplicate and corrected file behavior is undefined |
| 68 | - late or missing files do not create actionable alerts |
| 69 | - downstream publish opens before the landing validations complete |
| 70 | |
| 71 | ## Verification |
| 72 | |
| 73 | - [ ] The feed contract defines format, arrival, and ownership expectations |
| 74 | - [ ] Landing, quarantine, and duplicate-handling rules are explicit |
| 75 | - [ ] Validation covers schema, completeness, and control totals where relevant |
| 76 | - [ ] Replay and corrected-file behavior are documented before go-live |
| 77 | - [ ] Failure handling and partner escalation paths are operationally clear |