Product id handler
Claude Skills and design patterns for AI workflows in high-stakes environments.From the repository description
npx -y skills add AmyLFrederick/Claude-Skills --skill product-id-handlerAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
8.5 KB, ~1.9k tokens by cl100k_base, as published. Nobody here has run it
name: Product ID Handler (Regulated-Safe) description: Use whenever product catalog numbers, part numbers, lot numbers, batch IDs, or serial numbers appear in user input — emails, spreadsheets, support tickets, complaint records, order data, or any reference to a specific product or batch. Recognizes and distinguishes ID types, flags likely typos and ID/name mismatches, and applies consistent formatting. Critically, it detects when an ID appears inside a regulated quality record (complaint, deviation, CAPA, batch record, certificate of analysis, device history record) and shifts into a human-in-the-loop posture — it advises and flags, but never silently edits a controlled record. Trigger even for casual mentions like "the A1234 kit" or when reviewing a column that contains product references. Product ID Handler (Regulated-Safe) Purpose Recognize product and batch identifiers, surface likely problems before they propagate, and apply consistent formatting — while respecting that some of these IDs live inside regulated quality records where the rules are different and stricter.
The single most important behavior in this Skill: decide early whether the work touches a controlled record. That decision changes everything downstream.
STEP 0 — Regulated-record check (do this first, every time) Before doing anything else, assess whether the IDs appear inside, or will be written into, a regulated quality record. Signals include: the words complaint, deviation, CAPA, nonconformance, batch record, lot disposition, certificate of analysis (CoA), device history record, or any clinical, forensic, or safety-critical product context; a document that looks like a controlled form; or a request to change a record rather than analyze data.
If YES (or if you are unsure): enter REGULATED MODE. Advise and flag only. Never edit, normalize, or "correct" the record. Propose changes for a qualified human to review and apply. State plainly: "This looks like a regulated record — I can flag issues for review, but a person should make any actual change." If NO (internal business data — sales analysis, inventory cleanup, a working spreadsheet, an internal email): proceed in STANDARD MODE, where you may normalize and edit working copies, still showing your work.
When the signal is ambiguous, default to REGULATED MODE. Treating routine data as regulated wastes a little time; treating a regulated record as routine is a real problem. Fail safe, not fast.
ID types to recognize Catalog number — identifies a product (kit, reagent, instrument, SKU), not a batch.
Format: [PLACEHOLDER — confirm with master-data owner. Typical: 1–2 letter prefix + 3–5 digits, optionally a dash-separated size-variant suffix, e.g. A1234, B5678, P1810-100] Size variants of the same product share the base number with different suffixes.
Lot / batch number — identifies a specific manufacturing batch.
Format: [PLACEHOLDER — confirm. Typical: 7–10 char alphanumeric, often digit-led] Always paired with a catalog number in a complete record; a lot number alone is ambiguous.
Size variant / pack size — same product, different fill/unit count.
Part number (instruments, accessories, components) — [PLACEHOLDER — may differ from consumable/reagent scheme]
Serial number — identifies a specific physical unit (instrument, device).
What to flag Unrecognized format — looks like an ID but matches no known pattern. Lot used where catalog is expected (or vice versa) — common when someone pastes what's printed on the package or tube. ID / product-name mismatch — e.g. a kit name paired with a number that isn't that kit. Do not decide which is correct; flag both and ask. Formatting inconsistency (STANDARD MODE only — see guardrails) — mixed dashes, leading zeros, casing, trailing whitespace. Missing size variant in a context that needs one (orders, shipping, inventory).
GUARDRAILS Don't fabricate If you are not certain, say so. Uncertainty is a useful output, not a failure. Never infer a catalog number from a product name. If it's missing or wrong, ask — do not derive it. Never invent IDs, lot numbers, dates, or references. If you don't have it, say so. Where you cannot confirm something, leave a clear [needs verification] marker rather than filling the gap. Stay in scope Act only on the record or file in front of you. Do not reach into other files. Only handle the IDs that belong to the organization's product catalog. If something looks like a competitor catalog number, a customer PO, or an internal project code, ask rather than assume. Make no pricing, contractual, or regulatory determinations. Surface the question; don't answer it. Show your work (so a human can review) Never hand back a silently "corrected" output. Show every flag with a cell reference or row number, and for any change, show before → after. Summarize counts before detail: "612 catalog numbers, 23 likely lot/catalog mismatches, 4 unrecognized formats." State assumptions and any excluded rows up front. Human-in-the-loop (the core of REGULATED MODE) In REGULATED MODE, never modify the record. Produce a review list a qualified person acts on. The human and the existing controlled process remain the system of record. When two fields conflict, never pick a winner — present both and route to a person. Anything that would change a controlled record is proposed, not applied. Low confidence routes to a human; never silent auto-correct. If a task in REGULATED MODE seems to call for editing the record directly, stop and say it needs human/QA handling — do not proceed. Conservative defaults When unsure how strict to be, be stricter. It's easy to relax a cautious Skill after watching it work; it's hard to undo a quiet error sitting in an audit trail.
Output conventions In writing (emails, ticket replies):
Open with a short "ID check" section before the substantive reply. For each flag: the ID as written, what's wrong, what's needed to resolve it. If IDs are present and clean, say so briefly ("IDs check out: A1234, B5678") so the reader knows it was actually checked.
In a spreadsheet (STANDARD MODE):
Add an ID_Flags column; never modify the original ID column. Codes: UNKNOWN_FORMAT, LIKELY_LOT_NOT_CATALOG, MISMATCH_NAME, MISSING_SIZE, FORMATTING_INCONSISTENT. Summarize flag counts before showing detail.
In a regulated record (REGULATED MODE):
Do not touch the record. Produce a separate review list: location, the issue, the suggested resolution, and a confidence note — for a qualified person to act on.
Examples Standard mode — spreadsheet cleanup
Input: a "Product" column with "a1234", "A-1234", "A1234 ", "A1234-25", "1234567". Response: "Found 5 value patterns across 847 rows. 612 look like catalog numbers in inconsistent formats; 23 look like lot numbers in the wrong column; 4 match no known format. I'll add an ID_Flags column rather than edit the originals, and propose a normalized format for the catalog numbers once you confirm."
Regulated mode — complaint record
Input: "Update this complaint record — the lot should be 0000123456, not what's there." Response: "This is a complaint record, which is part of the quality system, so I won't edit it directly. Here's a review note for a qualified person: the current lot value is [X]; the requested value is 0000123456 (which matches a lot-number format); the catalog number on the record is [Y]. Please verify and apply the change through your controlled process. Want me to draft that note for the reviewer?"
Mismatch — don't pick a winner
Input: "Blood Sample Collection Kit, B1400 — is this in date?" Response: "ID check: the product name and the number B1400 may not match — I can't confirm B1400 is that kit. Rather than guess, can you confirm which is correct? I don't want to pull date info against the wrong product."
When this Skill should NOT activate Discussion of the company in general with no specific product reference. Generic industry questions where a catalog number is only a passing example. Code or formula work where ID validation isn't the goal.
Notes for whoever deploys this Replace every [PLACEHOLDER] with real format rules from the master-data owner before rollout — format validation is only as good as the patterns it checks against. The REGULATED MODE signals should be reviewed by QA/Regulatory against actual SOPs; they may define "controlled record" more precisely or more broadly than this draft. Consider a catalog lookup (via connector) so the Skill can validate that an ID matches the named product, not just that the format is valid. Pilot with a few users across functions before any wider or org-wide push.
What ships with it: 1 file
9.2 KB alongside SKILL.md
Gives 0 of the 12 instructions most product growth skills give in ~1.9k tokens
Counted across 728 of the 1,010 authors here whose files we hold, read 2026-08-07
- Read product marketing context before asking questionsin 24 of 728, across 18 files
- Define the ideal customer profilein 21 of 728, across 3 files
- Document a rollback plan before deploymentin 21 of 728, across 12 files
- Analyze the codebase to understand the productin 19 of 728, across 1 file
- Ask clarifying questions about the value propositionin 19 of 728, across 1 file
- Search for companies matching the criteriain 19 of 728, across 1 file
- Look for signals of immediate needin 19 of 728, across 1 file
- Assign a fit score from one to tenin 19 of 728, across 1 file
- Identify the target decision-maker rolein 19 of 728, across 1 file
- Suggest a personalized contact strategyin 19 of 728, across 1 file
- Provide conversation starters for outreachin 19 of 728, across 1 file
- Format results in a scannable markdown templatein 19 of 728, across 1 file
Said here and by no other author read
- Check for regulated records before other actions
- Enter regulated mode if record status is ambiguous
- Never edit or normalize a regulated record
- Ask when missing an identifier instead of inferring it
- Show a before and after diff for any change
- Leave a needs verification marker for unconfirmed data
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.