feat(github): auto-extract issue prompt on assignment and opened events

[dzianisv]

Summary

• Auto-extract issue title, body, and labels as the prompt when issues events fire with assigned or opened actions
• Removes the hard requirement for PROMPT env var on these issue actions
• Enables a "assign to bot → it works on it" workflow

Changes

packages/opencode/src/cli/cmd/github.ts

The getUserPrompt() function previously required the PROMPT environment variable for all issues events. Now:

1. If PROMPT is set, it's used as an override (backward compatible)
2. For assigned and opened actions: auto-constructs prompt from issue.title, issue.body, and issue.labels
3. For other actions (e.g. labeled, closed): still requires PROMPT

The downstream issue handling flow (branch creation, work, PR/comment) is unchanged.

Testing

• The change is self-contained in getUserPrompt() and doesn't affect any other code paths
• PROMPT env var override still works for all event types (backward compatible)
• The assigned and opened paths construct the same prompt format that PROMPT users would provide manually

Refs #19926

[github-actions]

This PR doesn't fully meet our contributing guidelines and PR template.

What needs to be fixed:

• PR description is missing required template sections. Please use the PR template.

Please edit this PR description to address the above within 2 hours, or it will be automatically closed.

If you believe this was flagged incorrectly, please let a maintainer know.

[dzianisv]

Test Results

Extracted the issue prompt building logic into a pure, exported buildIssuePrompt() function and added 11 unit tests covering:

Test Case	Status
Title + body formatting	✅
Null body fallback	✅
Undefined body fallback	✅
Empty string body fallback	✅
String labels array	✅
Object labels array ({ name })	✅
Mixed label types (string + object)	✅
Empty labels array (omit label line)	✅
Undefined labels (omit label line)	✅
Partial/invalid labels filtered out	✅
Markdown body preservation	✅

Run output

bun test v1.3.11 (af24e281)

 28 pass
 0 fail
 36 expect() calls
Ran 28 tests across 1 file. [3.86s]

All 28 tests pass (17 existing extractResponseText/formatPromptTooLargeError + 11 new buildIssuePrompt).

Typecheck also passes across all 13 packages in the monorepo (confirmed by pre-push hook).

[github-actions]

This pull request has been automatically closed because it was not updated to meet our contributing guidelines within the 2-hour window.

Feel free to open a new pull request that follows our guidelines.

[dzianisv]

E2E Test Results — Issue Assignment Trigger

Summary

The issue assignment trigger works end-to-end. When an issue is assigned in a repo with the opencode workflow, opencode automatically picks up the issue, extracts the prompt from the issue title/body (via buildIssuePrompt()), sends it to the LLM, and posts a response comment.

Test Setup

• Test repo: dzianisv/codebridge-test
• Workflow: .github/workflows/opencode.yml — triggers on issues: [assigned, opened]
• Model: opencode/gpt-5-nano (free tier, no API key required)
• Branch tested: feat/github-issue-assignment from dzianisv/opencode fork (runs opencode from source via bun run --cwd packages/opencode --conditions=browser ./src/index.ts github run)

Test Issue

• Issue: dzianisv/codebridge-test#683 — "Create a simple greeting script"
• Assigned to: dzianisv

Workflow Run

• Run: 23722794041 — SUCCESS ✅
• Duration: 51 seconds total

Key Log Lines (from the successful run)

Configuring git...
Asserting permissions for user dzianisv...
  permission: admin
Adding reaction...
opencode session ses_2c3dad884ffetyy7cCiVvERWqt
Checking out new branch...
Fetching prompt data for issue...
Sending message to opencode...

Implemented the simple greeting script as requested.
...
Creating comment...
Removing reaction...

Result

The opencode-agent GitHub App posted a comment on issue #683 with:

• A complete greet.py implementation matching the issue request
• A link to the opencode session
• A link to the GitHub Actions run

What This Proves

1. buildIssuePrompt() correctly extracts issue title + body into a prompt
2. getUserPrompt() no longer requires the PROMPT env var for issues.assigned / issues.opened events
3. The full pipeline works: issue assignment → workflow trigger → prompt extraction → LLM response → comment posted
4. No OPENCODE_API_KEY or other secrets are needed (uses free model with public API key)

Previous Failure (for comparison)

• Run: 23718780671 — failed with ProviderModelNotFoundError because opencode/claude-sonnet is not a valid model ID (correct ID: opencode/claude-sonnet-4)

[dzianisv]

E2E Test Results — ALL CHECKS PASS ✅

Test Setup

• Test repo: dzianisv/codebridge-test
• Test issue: #685 — "Add a calculator.py script with basic math operations"
• Workflow runs: 23723123435 (assigned) + 23723123200 (opened)
• Model: opencode/gpt-5-nano (free, no API key)

Fix Applied

• Commit 68abd7556: GithubRunCommand.handler() now uses process.env.GITHUB_WORKSPACE || process.cwd() to determine the target repository directory
• This ensures opencode operates on the correct repository checkout in GitHub Actions, regardless of where bun is invoked from

Full Flow Evidence (run 23723123435)

1. ✅ Workflow triggered on issues.assigned event
2. ✅ Prompt extracted from issue title+body automatically (no /opencode comment needed)
3. ✅ LLM processed the prompt: generated calculator.py with add/subtract/multiply/divide
4. ✅ Branch pushed: opencode/issue685-20260330004524 to test repo
5. ✅ PR created: #686 — "Added calculator.py CLI tool" (57 lines added)
6. ✅ Comment posted on issue conversation hangs if a large file is in the directory context #685 with PR link
7. ✅ No API key required — used free opencode/gpt-5-nano model

Acceptance Criteria

#	Check	Status
1	Workflow triggers on issues.assigned	✅ PASS
2	buildIssuePrompt() extracts title+body	✅ PASS
3	LLM processes the prompt	✅ PASS
4	Comment posted on issue	✅ PASS
5	Branch created and pushed	✅ PASS
6	PR created	✅ PASS
7	No API key required	✅ PASS

All 7/7 acceptance criteria passed in the same evaluation run.