Policy Retrieve
Purpose
Bring the governing policy text and config values into the case with citations, so the Govern step can quote them.
When to use
- After return-intake in every case.
- Ad hoc policy questions for a category ("what's the return window for electronics").
When NOT to use
- Making the determination or surfacing fraud signals - eligibility-check.
- Structuring a new request - return-intake.
- Drafting the packet or customer wording - case-packet.
- Warranty or price-match questions: this plugin ships return rules only (references/returns-rules.md); say so and escalate to the policy owner.
Where the data comes from
This plugin ships no sample data. The policy text (references/returns-rules.md) and windows/limits (config/return-windows.json) are retailer configuration inside this skill folder. The case itself is the user's own content:
case.jsonwritten earlier in the session by return-intake, or a case file the user names, links or attaches in the chat.- Their OneDrive or SharePoint, found with Work IQ search (Microsoft 365 search) by case ID or file name when no location is given; for an ad hoc policy question no case file is needed.
Read each source with the file reader (ReadFileContent) and cite it by file name. Write the updated case payload to the session working folder, never into the plugin. If the case file cannot be found, say so and ask the user where it is (or run return-intake first). Never invent case facts or fall back to example values.
Inputs
Configuration ships inside this skill folder (synced from the plugin-root shared/ copy before packaging): config/, contracts/, references/returns-rules.md - paths below are relative to this skill folder.
case.json (from return-intake in this session, or the user's case file - see Where the data comes from) + references/returns-rules.md + config/return-windows.json. Schema for the output block: policy_context in contracts/rtl.returns-refund-case.v1.json.
Steps
- Confirm the category exists in
config/return-windows.json.windows_days; if not, escalate to the policy owner and recordneeds_evidencecontext. - Select clauses matching the case shape: window (#1.1), receipt/order lookup (#1.2), condition (#1.3), signal rules present in the facts (#3.1-3.4), determination discipline (#2.1, #2.2), human gate (#4.1, #4.2), exceptions (#5.1).
- Write
policy_contextexactly per the contract:clauses[{clause_id, text (verbatim), citation, applies_because}],config_values[{key, value, unit, citation}],not_found[]for anything requested that the policy text does not contain,source: skill:policy-retrieve/1.0.4. - Validate the payload against the contract; write it back as the case payload (
case.jsonin the session working folder).
Example
User: "What does the return policy say for electronics with no receipt on a $120 item?"
-> read config/return-windows.json: windows_days.electronics = 15, receiptless_limit = 50.0
-> read references/returns-rules.md: #1.1 window clock, #1.2 receipt/order lookup, #3.3 high-value receiptless
-> policy_context.clauses = [RET-1.1, RET-1.2, RET-3.3 verbatim]; not_found = ["warranty"] if asked
Reply: table of clause id | verbatim text | applies because; then "Determination: not made here - eligibility-check."
Output format
The case payload with a policy_context block added and nothing else changed. In chat: a short list of the clauses attached (id + one line) and the config values, then "Determination: not made here - eligibility-check".
Grounding requirements
Clause text quoted verbatim with ids; config values cite config/return-windows.json. If a clause does not exist, it goes in not_found - it is never written.
Guardrails
- Retrieval only; the determination belongs to eligibility-check. Never mark a case eligible/ineligible here.
- Never paraphrase, soften or "round" a window or limit; attach the configured value exactly.
- Never invent a clause or clause id for policy areas the shipped text does not cover.
Escalation / uncertainty
If the policy file or config cannot be read, stop and report the missing source; never quote a clause from memory. If a category is missing from config, escalate to the policy owner. Category missing from config, or a requested policy area absent from the text: escalate to the policy owner; the case becomes needs_evidence.