fix(cli/init): rephrase agent prompt with ASD-STE100

[Xe]

Agents need deterministic phrasing to behave predictably. This patch is a further round of hardening to ensure that agent instructions result in more deterministic output.

I applied the concepts from ASD-STE1001 to ensure that there is one instruction per sentence, active voice for every action, and that conditions prefix sentences.

The goal is to make instructions unambiguous for AI agents. A less capable model reads these instructions the same way as a more capable model.

Ref: #230

---

Note

Low Risk
Copy-only changes to agent-facing prompt text and matching tests; no CLI execution or credential-handling logic is modified.

Overview
tigris init --agent now prints a setup recipe whose wording follows ASD-STE100-style rules: imperative sentences, conditions before commands, and compound steps split so each sentence is one action.

In buildAgentSetup (plan.ts), every step string is rephrased without changing the underlying workflow—install/update lead-in, auth, org selection, bucket creation, access-key temp file handling, .env scripting, and closing links/agent-config snippet. The intro line becomes “Help the user configure Tigris for this project” instead of “set up their project with Tigris.” Secret-handling guidance is spelled out as explicit bullets (script reads credentials; agent must not load the secret into context).

Tests in plan.test.ts are updated to assert the new step-one and step-two phrasing. A changeset documents the patch-level CLI behavior change for consumers.

^{Reviewed by Cursor Bugbot for commit c2fe7c8. Bugbot is set up for automated code reviews on this repo. Configure here.}

[greptile-apps]

Greptile Summary

The PR rewrites the tigris init --agent setup recipe in simplified, imperative language to make agent execution more deterministic.

• Splits compound setup instructions into shorter sentences with explicit conditions and actions.
• Updates install, authentication, bucket, access-key, environment, and documentation guidance.
• Adjusts the init-plan tests and adds patch changesets for both CLI packages.

Confidence Score: 4/5

The organization-selection instruction should be fixed before merging because agents can execute it without a concrete organization and stop the setup flow.

The rewritten condition governs only asking the user, while the required orgs select <org> command is presented as the next unconditional action and fails when no organization value is available.

Files Needing Attention: packages/cli/src/lib/init/plan.ts

Important Files Changed

Filename	Overview
packages/cli/src/lib/init/plan.ts	Rephrases the generated agent recipe, but separates organization selection from its intended condition and can direct agents to run it without a resolved organization.
packages/cli/test/lib/init/plan.test.ts	Updates assertions for the revised wording but does not cover the conditional organization-selection instruction.
.changeset/cli-init-agent-recipe-simplified-english.md	Adds appropriate patch release notes for the agent-recipe wording changes.

_{Reviews (1): Last reviewed commit: "fix(cli/init): rephrase agent prompt wit..." | Re-trigger Greptile}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Want higher recall? High effort reviews run extra passes and find more bugs. A team admin can switch effort levels in the Cursor dashboard.

^{Reviewed by Cursor Bugbot for commit c2fe7c8. Configure here.}

[cursor]

Org select lost its condition

Medium Severity

Splitting the org step into STE sentences left tigris orgs select without the multi-org condition that used to gate it. Agents can now treat select as always required even when <org> was never chosen, which is common for single-org users and undermines the deterministic phrasing this change aims for.

 

^{Reviewed by Cursor Bugbot for commit c2fe7c8. Configure here.}

[greptile-apps]

Organization selection loses its guard

When the account has zero or one organization, the condition only guards asking the user, while the separate Then run instruction still directs the agent to execute tigris orgs select <org>. Without a concrete organization, this required positional argument produces an organization-required or organization-not-found error and stops setup.

Suggested change

	'Run `tigris orgs list --format json`. If there is more than one org, ask the user which org to use. Then run `tigris orgs select <org>` to make that org active. The commands that come after target the active org.',
	'Run `tigris orgs list --format json`. If there is more than one org, ask the user which org to use. If there is more than one org, run `tigris orgs select <org>` to make that org active. The commands that come after target the active org.',

Knowledge Base Used: Tigris CLI