Dx Org Manage

MANDATORY: Follow these instructions exactly. Do NOT fall back to MCP tools.

Tool constraint: Use the Bash tool for all sf CLI commands. Always include --json for structured output. Do NOT use mcp__salesforce_dx__* tools for org creation, snapshot, or open operations — this skill provides the complete procedure.

Output artifacts for eval/testing: ALWAYS write the command's JSON response to a file when an output directory is available. Do NOT ask the user what file to write — this skill defines the filenames. After executing the command: (1) if the user specified an output path (e.g. "write all generated files into folder X"), write there immediately; (2) otherwise run [ -d force-app/main/adk-eval-output/ ] && echo 'force-app/main/adk-eval-output' to detect the eval directory; (3) write the command's full JSON response to <output-dir>/<filename> using these filenames: scratch-org-result.json for org creation (for a batch of N orgs, scratch-org-result-1.json … scratch-org-result-N.json), scratch-org-list-result.json for list, org-display-result.json for display, scratch-org-resume-result.json for resume, or scratch-org-delete-result.json for delete. This is the generated output — write it without asking. (Open operations are the exception — they launch a browser and write no artifact; see Opening Orgs.)

Creating Scratch Orgs

REQUIRED steps — execute in order:

Step 1. Resolve the environment and creation method:

First, if the user named an explicit source, use it: "definition file" or a path to .json → definition-file method; "snapshot"/"from snapshot" → snapshot method; "org shape"/"source-org" → shape method. Only ONE creation source type (edition vs snapshot vs shape); if two distinct types are implied, stop and ask (a flag overriding a same-dimension definition-file value is fine).

Otherwise resolve by environment:

• AUTO MODE (zero-prompt): if sfdx-project.json is present AND exactly one config/*scratch-def.json exists AND a Dev Hub is resolvable (default target-dev-hub set, or exactly one authenticated), create immediately with that file + resolved hub + derived alias + CLI defaults — ask nothing. This also honors a batch count (see below).
• STATE A (in a project), not auto: enumerate config/*scratch-def.json — 0 files → default --edition developer (or author one if features/settings are wanted, see below); 1 file → use it; many → ask which one (never silently pick).
• STATE B (no sfdx-project.json): do NOT block and do NOT silently create a throwaway — present the guided 3-way choice (point to a project / scaffold one via sf template generate project --name <name> / create a throwaway here, then guide the source). See references/scratch-org-create.md.

Batch — "create N scratch orgs": there is no native count flag — loop the create command N times with N distinct aliases (<base>-1 … <base>-N, each collision-guarded so it never re-points an existing org). Report each org and write one artifact per org. On a mid-loop Dev Hub limit error, surface the CLI's error unchanged and report which orgs already succeeded.

Definition-file authoring: if features/settings are requested and no suitable file exists, author a new purpose-named definition file (seed-then-modify; documented fields only; no-clobber), show it, and create from it. In a non-interactive/eval context, author the file AND proceed to create without waiting for edit confirmation. Details in references/scratch-org-create.md and references/definition_file_options.md.

Step 2. Resolve the Dev Hub to a concrete value — then pass it explicitly. Resolve it once to an actual username or alias and pass that exact value via --target-dev-hub on every command in Step 3. Resolution order:

1. A Dev Hub the user named explicitly → use it verbatim.
2. Else the default: the non-empty result[0].value from sf config get target-dev-hub --json.
   bash

   sf config get target-dev-hub --json

3. Else the single authenticated Dev Hub. Run this exact command — do NOT hand-write your own filter. A Dev Hub can appear in any of the sf org list buckets (devHubs, nonScratchOrgs, other, sandboxes, scratchOrgs); a filter that checks only one bucket (e.g. only .other[]) misses it and makes you think no hub exists:
   bash

   sf org list --json | jq -r '[.result.devHubs[]?, .result.nonScratchOrgs[]?, .result.other[]?, .result.sandboxes[]?, .result.scratchOrgs[]?] | map(select(.isDevHub == true).username) | unique | .[]'

   • Exactly one username printed → use it.
   • Zero printed → no hub is authenticated. Do NOT run sf org create at all — there is nothing to pass to --target-dev-hub, and any create attempt will fail. Instead, STOP here: advise sf org login web --set-default-dev-hub, and if an output directory is available write that advisory as the artifact. Do not proceed to Step 3.
   • Two or more → ask the user which one (do not pick arbitrarily).

• Never invent or guess a Dev Hub name. Fabricating a placeholder alias (e.g. eval-target, my-dev-hub, DevHub) is the top eval failure — the CLI correctly rejects it with NotADevHubError. If the command above prints nothing, that means no Dev Hub exists in this environment — it does NOT mean you should substitute a name. There is no valid fallback name to invent: an unresolved hub is a hard stop, not a value to guess. Do NOT run sf org create scratch with a made-up --target-dev-hub, and do NOT run it with no --target-dev-hub flag either (that yields NoDefaultDevHubError). Stop and advise sf org login web --set-default-dev-hub.
• The default target-dev-hub may be directory-scoped in some CLI setups (sf config get can return empty after a cd), which is why step 3's all-bucket sf org list check is the reliable fallback — it is not directory-scoped.
• Do NOT proceed until a concrete Dev Hub value is resolved.

Step 3. Build and execute the command based on method:

Definition file:

sf org create scratch --definition-file <path> --target-dev-hub <alias> --alias <name> --json

Edition only:

sf org create scratch --edition developer --target-dev-hub <alias> --alias <name> --json

From snapshot:

sf org create scratch --snapshot <snapshot-name> --target-dev-hub <alias> --alias <name> --json

From org shape:

sf org create scratch --source-org <source-org-id> --target-dev-hub <alias> --alias <name> --json

--source-org takes the 15-character source org ID — the ID of the org the shape was captured from (a 00D… org ID), NOT the 3SR… shape record ID shown by sf org list shape. Pass the ID the user gives you through unchanged. If the CLI rejects it (e.g. InvalidIdLengthError, InvalidPrefixError), surface that error verbatim and stop — do NOT truncate, reformat, guess, or retry with a different ID.

Apply these flags when requested:

• --duration-days <days> — default 7, max 30
• --set-default — make this the default org
• --no-track-source — disable source tracking (for CI/CD)

Step 4. MANDATORY - Run org list and write output (SUCCESS path only): This step runs only when Step 3 created an org successfully. If Step 3 returned an error, SKIP this step and follow Error handling below instead. After the org is created, you MUST run this command:

sf org list --json

Then:

1. Parse the JSON result and find the scratchOrgs array

2. Find the entry where username matches the username from Step 3's creation result

3. Extract that complete org object (it will include: alias, username, orgId, instanceUrl, loginUrl, isDefaultUsername, orgEdition, status, expirationDate, devHubUsername, and other fields the CLI returns).

4. Report to the user:

   • Created scratch org.
   • Alias: [alias from the org list entry]
   • Username: [username]
   • Org ID: [orgId]

5. If an output directory is available (per the output artifacts rule above), write that extracted org object to <output-dir>/scratch-org-result.json as-is — this skill is a pass-through wrapper. Write every field the CLI returns for that org; do NOT curate, whitelist, or drop non-secret fields (e.g. instanceName, createdOrgInstance, signupUsername, orgName, edition are non-secret metadata the CLI returns freely — keep them). The one thing you never emit is a live secret: accessToken and sfdxAuthUrl — and the CLI already redacts these in --json output (they arrive as "[REDACTED] …"), so simply preserve that redaction and never un-redact or re-derive the real value.

Example: If sf org list --json returns {"result": {"scratchOrgs": [{"alias": "feature-dev", "username": "[email protected]", "orgId": "00D...", ...}]}}, write the inner org object {"alias": "feature-dev", "username": "[email protected]", "orgId": "00D...", ...} — the full object for that org — to the file.

Write the extracted org-list entry (the resolved org record), NOT the raw creation-command response. Do NOT suggest verification steps to the user.

Error handling (Step 3 create failed — no org was created): Surface the CLI's error output to the user verbatim (no rewriting, no retry, no editing the definition file). Do NOT run Step 4 (there is no org to list/extract). If an output directory is available, write the create command's raw error JSON as-is to <output-dir>/scratch-org-result.json — this is the command's response and is the output artifact for the failed run. (For a batch, write the error to the artifact for the org that failed, and still write the success entries for any orgs that already succeeded before the failure.) Then, where the CLI's error matches one of these, add the corresponding pointer:

• "Snapshot not found" → suggest sf org list snapshot --target-dev-hub <alias>
• "No default Dev Hub" → advise sf org login web --set-default-dev-hub

When you need more detail:

• For the complete creation workflow (AUTO MODE, STATE A/B, batch, definition-file authoring) → load references/scratch-org-create.md; for list/display/resume/delete → load references/scratch-org-operations.md
• For available features, settings, and definition file structure → load references/definition_file_options.md
• For edition selection guidance and comparison → load references/edition_types.md
• To create/check/list/delete the snapshot itself before referencing it here → use the dx-org-snapshot-manage skill

Listing Scratch Orgs

Step 1. Execute:

sf org list --json

Step 2. Report + write output: default view = active scratch orgs from result.scratchOrgs[]; report alias / username / orgId / expiration per org. Document --all (include expired/deleted) and --clean as options, not the default. Write the result.scratchOrgs[] array to <output-dir>/scratch-org-list-result.json (per the output-artifacts rule).

Detail: references/scratch-org-operations.md.

Displaying an Org

Step 1. Execute:

sf org display --target-org <alias> --json

Step 2. Report + write output: report alias / username / orgId / instanceUrl / status / expiration. Write the wrapped {status, result} JSON to <output-dir>/org-display-result.json — do NOT unwrap. In any committed example/gold, accessToken and sfdxAuthUrl must be redacted (value begins with [REDACTED]).

--verbose: never add it and never run it from this skill — it returns sfdxAuthUrl (a refresh token) into agent context. If the user needs the auth URL, tell them to run sf org display --verbose themselves in their own terminal.

Detail: references/scratch-org-operations.md.

Resuming Scratch Org Creation

For a create that ran with --async or timed out (exit code 69).

Step 1. Execute:

sf org resume scratch --job-id <id> --json

If the user gave an explicit --job-id, use it; otherwise default to --use-most-recent. If there is no recent job, surface the CLI's not-found result unchanged and point the user to sf org list — do not invent a job id.

Step 2. Write output: write the command's JSON as-is to <output-dir>/scratch-org-resume-result.json.

Detail: references/scratch-org-operations.md.

Deleting a Scratch Org

Destructive — no undo.

Step 1. Confirm before running. Ask "Delete scratch org X?" and wait for confirmation, UNLESS the user already gave explicit deletion intent. If the target is the current default org, call that out in the confirmation (extra-guard).

Step 2. Execute (only after confirmation):

sf org delete scratch --target-org <alias> --no-prompt --json

--no-prompt is passed after the skill's own confirmation, so the agent isn't left waiting on the CLI's interactive prompt.

Step 3. Write output: write the command's JSON as-is to <output-dir>/scratch-org-delete-result.json.

Detail: references/scratch-org-operations.md.

Opening Orgs

REQUIRED steps — execute in order:

Browser launch only — never --json, never --url-only. Open is the one operation in this skill that writes NO artifact. Plain sf org open launches the browser and logs the user in without ever printing a login URL, so no credential enters agent context or a file. Do NOT add --json: sf org open --json returns a live login URL in result.url (/secur/frontdoor.jsp?otp=/sid=<token>) that is credential-equivalent — pulling it into context (even to redact it before writing) is the exact S1 leak the plain command avoids. Same reason --url-only is banned: if the user explicitly asks for the URL and not a browser (e.g. "URL only", "just give me the link", "headless/remote"), do NOT run it — tell them to run sf org open --url-only themselves in their own terminal (same handling as sf org display --verbose). A vague "open my org" just opens the browser.

Step 1. Match user request to command:

Step 2. Execute the matching command using the Bash tool. Plain sf org open prints nothing on success — that is expected.

Step 3. Report result: Report to the user that the org (or path/metadata file) was opened in the browser. There is NO artifact to write for open operations — do not write org-url-result.json or any file, and do not add --json to produce one. If the command errors, surface it per Error handling below.

Error handling:

• "no target org" → advise sf config set target-org <alias>
• "auth error" → advise sf org login web --alias <alias>

When you need more detail:

• For complete opening org workflow and all available flags → load references/opening-org.md

Reference File Index

Load these reference files for detailed guidance:

To create/check/list/delete a snapshot itself (rather than just consuming one), use the dx-org-snapshot-manage skill.

Example Files

Example command outputs for testing and troubleshooting: