Automation: Sandbox Post-Copy Config Generate

Convert a customer's sandbox-refresh / post-copy SOP into a structured JSON array that the post-copy automation tool consumes. Each entry is a declarative instruction: which Salesforce configuration to update, which fields are involved, whether it is active, and what order it runs in.

STOP — do this before writing any JSON

Do not compose the output from memory. Before you write the file, you MUST open and read assets/config_template.json and copy an entry from it for each action. Every output entry is exactly one of these two shapes — five top-level keys, no others, no wrapper object:

[
  {
    "ConfigurationName": "OutboundMessages",
    "Label": "IR_Account_OBM_PROD",
    "Fields": { "EndpointUrl": "https://uat.example.com/services/account", "Object": "Account" },
    "IsActive": true,
    "ExecutionOrder": 1
  },
  {
    "ConfigurationName": "RemoteSiteSettings",
    "Label": "R12_Remote_Site",
    "Fields": { "RemoteSiteUrl": "https://uat.example.com" },
    "IsActive": true,
    "ExecutionOrder": 2
  }
]

• ConfigurationName: exactly OutboundMessages or RemoteSiteSettings — never Type, Name, or Operation.
• OBM Fields: EndpointUrl + Object (both required). RemoteSite Fields: RemoteSiteUrl only — never Url/RemoteSiteURL.
• Top level is a JSON array. No steps/actions/records wrapper. No <…> or REPLACE_WITH_… placeholder ever survives into the output.

If you announce "I will now write …" without having read the template and catalog, stop and read them first — a from-memory guess produces the wrong keys and fails at runtime.

Scope

• In scope: Reading a customer SOP in any of the supported formats (PDF, xlsx, csv, JSON, docx, Markdown, plain text, pasted excerpt, or images containing data tables — e.g., a screenshot of an Outbound Messages list with endpoint URLs), identifying post-copy / post-refresh actions, mapping each action to a supported ConfigurationName, emitting the canonical JSON array.
• Out of scope: Generating Salesforce metadata XML (delegate to generating-* skills), deploying anything to an org, running the post-copy tool, inferring or fabricating values not present in the SOP — if the SOP does not give a concrete URL/value for an action, skip the action.

Every emitted entry must have every Field populated with a real value from the customer source. No empty strings, no null, no <from-backup> / TBD / TODO placeholders. Customers should never see an unpopulated field in the output — if a value cannot be located, skip the entry and surface it. See the corresponding rule below.

---

Required Inputs

Gather or infer before generating:

• SOP source(s): One or more paths (or pasted content) in any of: PDF, xlsx, csv, JSON, docx, Markdown, plain text, or images (.png/.jpg/.jpeg/.tiff/.bmp). Multiple files are common — the action list and the endpoint table sometimes live in different files. Read every file the user supplies.
• Target output path: Where the JSON config should be written. Default to post-copy-config.json in the current directory unless specified.
• Scope filter (optional): If the SOP covers many environments (e.g., fcQA, fcUAT, multiple sandboxes), confirm which subset the user wants in the output.

If the user provides a clear SOP and target, generate immediately without asking unnecessary questions.

---

Workflow

All steps are sequential. Steps 1–5 (reading the SOP, the catalog, the template, and the schema) are prerequisites to writing — you may not skip to the write step. If you catch yourself about to emit JSON without having read assets/config_template.json and references/configuration_catalog.md, go back and read them first.

1. Locate and read every supplied SOP source — read references/source_format_handling.md for the exact extraction recipe per format (PDF, xlsx, csv, JSON, docx, image). At a glance:

   • PDF: extract text with pypdf (text layer) and OCR image-based pages with pytesseract if the text layer is empty.
   • xlsx: read every sheet with openpyxl (data_only=True), scan all columns including ones outside the visible default range, check cell comments and embedded media.
   • csv / JSON / Markdown / text: read directly.
   • docx: extract paragraphs and tables with python-docx.
   • Images (.png/.jpg/...): use the Read tool to view, then decide if the image carries data (a table of endpoint URLs, a setup screenshot showing values to capture) or is purely illustrative (architecture diagram, flow chart). Extract values only from data-bearing images. See the image-handling rules in references/source_format_handling.md.
   • For very large SOPs (>50 pages / >20 sheets), focus on sections or sheets titled "Post Refresh", "Post-Copy", "Post-Refresh Steps", "Update …", or equivalent.

2. Identify post-copy actions — read references/sop_parsing_patterns.md for the heuristics that turn prose instructions ("Update Outbound Message endpoint X to URL Y") into structured action records.

3. Map each action to a ConfigurationName — load references/configuration_catalog.md. The catalog currently supports only OutboundMessages and RemoteSiteSettings. Any action that targets a different configuration type is out of scope: skip it and list it in the response so the user can extend the catalog later.

4. Read the JSON template — load assets/config_template.json. It shows the exact required shape of one OutboundMessages entry and one RemoteSiteSettings entry, with <…> placeholder slots. Copy an entry, replace every <…> slot with the concrete SOP value, and keep the exact top-level keys (ConfigurationName, Label, Fields, IsActive, ExecutionOrder) — never rename them to Type, Name, Operation, etc. Never emit an entry that still contains a <…> placeholder; if you cannot fill a slot, skip the entry (see Rules).

5. Validate against the schema — load assets/json_schema.json. Every entry must conform: ConfigurationName is one of the catalog values, Fields is an object, IsActive is boolean, ExecutionOrder is a positive integer.

6. Group entries by phase, then assign ExecutionOrder — ExecutionOrder is a phase number, not a per-row counter. Entries that can run in parallel (no dependency between them) share the same value. Different ConfigurationName types typically get different phases; all entries within one phase share its number. See the ordering heuristic in references/sop_parsing_patterns.md.

7. Compare against the example — verify the output shape against examples/sample_sop_to_config.json before writing.

8. Write the JSON file — emit pretty-printed JSON (2-space indent).

---

Rules / Constraints

Constraint	Rationale
Output is a JSON array at the top level (not an object with a wrapper key)	The post-copy tool consumes an array directly
Every entry has all five required keys: ConfigurationName, Label, Fields, IsActive, ExecutionOrder	The tool fails fast on missing keys; partial entries are not silently accepted
ConfigurationName is one of the catalog values	Unknown values cause the runtime mapper to error out — never invent a new type without updating the catalog
Fields keys are the actual API field names on the target metadata	Wrong key names mean the tool can't locate the field at runtime
Fields values are the concrete values from the SOP (literal URLs, names, etc.)	The post-copy tool applies the value as-is; placeholders are not resolved at runtime
If the SOP names an action but no concrete value (URL, etc.) is provided, skip the entry entirely and list it in the response	Generating an entry without a real value would produce a silent no-op or a deployment error at runtime
Every Field in every emitted entry must be a real value sourced from the customer's SOP or a supplemental sheet they provided. Never emit "", null, or placeholder markers like <from-backup> / TBD / TODO	Customers consume the JSON directly — empty / placeholder fields surface to them as broken output and would also fail at runtime
Before skipping an OBM / RemoteSite for missing values, search every tab / sheet of the supplied workbook (and every supplied file) for an endpoint table keyed by that name	Customer SOPs frequently split the action list and the URL table across different sheets (e.g., the Michelin UAT Refresh Planner lists OBMs in the Integration tab but the URL table lives in the Evolution SFA tab)
Information already captured by another field is not repeated in Fields (e.g., RemoteSiteName lives in Label, IsActive lives at the top level — neither belongs in Fields)	Duplicate keys make the entry ambiguous and waste bytes the tool then has to reconcile
For OutboundMessages, Fields MUST include both EndpointUrl and Object (the target SObject — Account, Contact, Asset, Lead, etc.)	Same Label can apply to multiple OBMs differing only by entity; Object disambiguates them at runtime
For RemoteSiteSettings, the URL key MUST be spelled exactly RemoteSiteUrl — never Url, RemoteSiteURL, SiteUrl, or EndpointUrl	The post-copy tool matches the field by exact API name; any other spelling means it can't locate the field and the entry silently no-ops at runtime
ExecutionOrder is a phase number — entries with no dependency on each other share the same value	The post-copy tool runs all entries with the same ExecutionOrder in parallel; sequencing is only needed where one action depends on another
IsActive: false entries remain in the output (do not delete them)	The customer toggles them on per-environment; deleting loses traceability
Default IsActive to true when the SOP marks an action as required	Most SOP steps are required; explicit opt-out is the exception

Canonical entry shape

See the entry shapes shown in the STOP section above, or copy directly from assets/config_template.json. Do not rename the five top-level keys (ConfigurationName, Label, Fields, IsActive, ExecutionOrder) to Type, Name, Operation, apiName, etc., and do not wrap the array in an object with a steps / actions / records key.

---

Gotchas

Issue	Resolution
SOP step says "delete all endpoints" rather than "update X to Y"	Emit an entry with ConfigurationName: "OutboundMessages", Label describing the deletion target, and a note in Fields ({"Action": "Delete"}); the catalog documents the Delete pattern
Same Label appears for multiple environments (fcQA + fcUAT)	Emit one entry per environment; suffix the Label or use the environment-specific Label as it appears in the SOP
For OutboundMessages, the same Label can legitimately apply to multiple entries when each targets a different Object (e.g., one OBM for Account, one for Contact).	Do not collapse them — emit one entry per Object. The Object value (inferred from the OBM name like IR_Account_OBM_PROD → Account) is what distinguishes them, not the Label.
SOP groups many Custom Labels together in a single table	One entry per row of the table — do not collapse into a single bulk entry
Action targets a setting outside the catalog (CustomLabels, ConnectedApps, NamedCredentials, SSO, CustomSettings, etc.)	Skip the action, do NOT invent a new ConfigurationName; list every skipped action in the response so the user knows what to add to the catalog later
SOP includes pre-refresh steps interleaved with post-refresh	Filter out pre-refresh — only post-refresh / post-copy actions belong in the output
SOP names an outbound message / remote site to update but does not include the new URL	Skip the entry. List the skipped item in the response so the user can supply the URL or amend the SOP
SOP includes secrets in URLs (e.g., https://USER:TOKEN@host/...)	Embed them verbatim — the tool consumes the value as-is — but flag in the response so the user is aware the JSON now contains a secret and should be stored / shared accordingly
User's SOP is in a non-English language or has heavy formatting (tables, callouts)	Extract the prose first, normalize whitespace, then parse — formatting artifacts do not need to survive into the JSON
The SOP includes screenshots / images. Some are illustrative (architecture diagrams, flow charts, "this is what the setup screen looks like" examples); others are data-bearing (a screenshot of a real Outbound Messages list, a captured Remote Site table).	Treat illustrative images as out-of-scope — do NOT extract values from them. For data-bearing images (the surrounding text refers to "the values shown below" / "as captured in the screenshot above" / a real Endpoint URL is visible), OCR them and use the values. See references/source_format_handling.md for the heuristic.
The SOP is supplied as multiple files (e.g., a PDF + a companion xlsx + a screenshot folder), or values are split across tabs / pages	Read every supplied file end-to-end before skipping any entry. The action list and the URL table are frequently in different files — see the rule above about multi-sheet sources.

---

Output Expectations

Deliverables:

• A single JSON file (default post-copy-config.json) containing a top-level array of post-copy action entries.
• A short summary in your response listing: total entries, count per ConfigurationName, actions skipped because they did not map to the catalog, and actions skipped because the SOP did not include a concrete value.

The output file structure conforms to assets/json_schema.json.

---

Cross-Skill Integration

Need	Delegate to
Generate the actual Salesforce metadata XML for a Custom Label / Named Credential / Remote Site / etc.	The matching generating-* skill
Deploy generated metadata or run the post-copy tool against an org	deploying-metadata skill
Compare an SOP against entities defined in UDD	udd:research-entities

---

Reference File Index

File	When to read
assets/config_template.json	Step 4 — starting structure for the JSON output
assets/json_schema.json	Step 5 — validate every emitted entry
references/configuration_catalog.md	Step 3 — to map an SOP action to a ConfigurationName and get the canonical Field keys
references/sop_parsing_patterns.md	Step 2 and step 6 — heuristics for extracting actions from prose and ordering them
references/source_format_handling.md	Step 1 — exact extraction recipe per input format (PDF / xlsx / csv / JSON / docx / image) and the data-bearing-vs-illustrative image heuristic
examples/sample_sop_to_config.json	Step 7 — verify output shape matches expected format
examples/sample_sop_excerpt.md	Step 7 — see a representative SOP excerpt that produced the sample JSON