Make prioritize agent multi-forge (GitHub + GitLab)

[ggallen]

Context

The fullsend platform is converting all agents from GitHub-only to multi-forge (GitHub + GitLab) support. The conversion follows a consistent pattern established by triage (merged), code, review, and retro (all in PR). Prioritize is Phase 6 in the rollout.

The plan on branch worktree-multi-forge-agents-plan in the fullsend repo (docs/plans/multi-forge-agents-work-breakdown.md) tracks the overall effort.

Decision: GitLab score storage — The forge_update_project_scores function on GitLab will attempt to write scores via the GitLab custom fields API (available on Premium/Ultimate). If the API returns 403/404 (feature unavailable on Free tier), it logs a notice and returns success — graceful degradation, not failure. The reasoning comment (which already contains the full score breakdown) is always posted regardless of tier, so no information is lost.

The Established Multi-Forge Pattern

Every converted agent follows the same 6-layer pattern. Prioritize must follow it exactly.

• Harness YAML — Remove top-level policy:. Add parallel forge.github / forge.gitlab blocks with: policy, skills, host_files, env (including FULLSEND_FORGE).
• Agent prompt — Forge-neutral. GITHUB_ISSUE_URL → ISSUE_URL. Replace gh CLI references with "use your forge skill". Add curl to tools.
• Env files — Split into env/github/<agent>.env and env/gitlab/<agent>.env. Map forge-specific vars to neutral names.
• Scripts — Shared pre/post scripts for both forges. Dispatch internally via ops library (<agent>-ops.lib.sh → github-<agent>-ops.lib.sh / gitlab-<agent>-ops.lib.sh). Both implement identical forge_*() function signatures.
• Policies — Split into policies/github/<agent>.yaml (allows gh binary) and policies/gitlab/<agent>.yaml (allows curl binary).
• Skills — Inject skills/github-forge or skills/gitlab-forge per forge.

Key: both forges share the same script files — dispatch happens inside via the ops library, NOT via separate script files per forge.

Changes

1. Agent prompt — agents/prioritize.md

• description: Score a GitHub issue → description: Score an issue
• tools: Bash(gh,jq) → tools: Bash(gh,curl,jq)
• "evaluate a single GitHub issue" → "evaluate a single issue"
• GITHUB_ISSUE_URL → ISSUE_URL, add examples for both forges
• Replace hardcoded gh issue view "$GITHUB_ISSUE_URL" --json ... → "Use the forge-appropriate command from your forge skill to fetch the issue"
• "A post-script handles all GitHub mutations" → "A post-script handles all forge mutations"
• Remove all other "GitHub" references in prose

2. Harness — harness/prioritize.yaml

• Remove top-level policy: policies/prioritize.yaml
• Update forge.github block to add:
  • policy: policies/github/prioritize.yaml
  • skills: [skills/github-forge]
  • host_files: entry for env/github/prioritize.env
  • Normalize env: ISSUE_URL: ${GITHUB_ISSUE_URL}, add FULLSEND_FORGE: github
  • Keep ORG and PROJECT_NUMBER (GitHub-only, for Projects V2)
• Add forge.gitlab block with:
  • policy: policies/gitlab/prioritize.yaml
  • pre_script: scripts/pre-prioritize.sh
  • post_script: scripts/post-prioritize.sh
  • skills: [skills/gitlab-forge]
  • host_files: entry for env/gitlab/prioritize.env
  • env.runner: — ISSUE_URL: ${GITLAB_ISSUE_URL}, GITLAB_TOKEN, GITLAB_HOST, FULLSEND_FORGE: gitlab
  • env.sandbox: — ISSUE_URL, GITLAB_TOKEN, GITLAB_HOST, FULLSEND_FORGE: gitlab

Use harness/triage.yaml as the structural template.

3. Env files

• Strip env/prioritize.env of forge-specific vars (remove GITHUB_ISSUE_URL and GH_TOKEN exports; keep if any shared vars remain, delete if empty)
• Create env/github/prioritize.env:

  export ISSUE_URL="${GITHUB_ISSUE_URL}"
  export GH_TOKEN="${GH_TOKEN}"
  export FULLSEND_FORGE="github"
  

• Create env/gitlab/prioritize.env:

  export ISSUE_URL="${GITLAB_ISSUE_URL}"
  export GITLAB_TOKEN="${GITLAB_TOKEN}"
  export GITLAB_HOST="${GITLAB_HOST}"
  export FULLSEND_FORGE="gitlab"
  

4. Policy split

• Move policies/prioritize.yaml → policies/github/prioritize.yaml (contents unchanged)
• Create policies/gitlab/prioritize.yaml — same structure but:
  • Replace github_api network policy with gitlab_api
  • Endpoints: gitlab.com and gitlab.cee.redhat.com port 443
  • Binaries: **/curl and **/node (not **/gh)
  • Keep vertex_ai policy unchanged

Use policies/gitlab/triage.yaml as the template.

5. Ops library (new files)

Follow the pattern from scripts/lib/triage-ops.lib.sh and forge-specific ops.

• Create scripts/lib/prioritize-ops.lib.sh — dispatcher:

  case "${FULLSEND_FORGE:-}" in
    github) source "${SCRIPT_DIR}/lib/github-prioritize-ops.lib.sh" ;;
    gitlab) source "${SCRIPT_DIR}/lib/gitlab-prioritize-ops.lib.sh" ;;
    *) echo "ERROR: invalid FULLSEND_FORGE" >&2; exit 1 ;;
  esac

• Create scripts/lib/github-prioritize-ops.lib.sh — extract current GitHub logic from post-prioritize.src.sh into forge_*() functions:

  • forge_validate_issue_url — current GitHub URL regex
  • forge_parse_issue_url — extract REPO, ISSUE_NUMBER from URL
  • forge_update_project_scores — current Projects V2 GraphQL logic (resolve project/item IDs, update 5 custom fields)
  • forge_post_comment — current fullsend post-comment logic via CSMA
  • Sources github-api-csma.lib.sh internally

• Create scripts/lib/gitlab-prioritize-ops.lib.sh — GitLab equivalents:

  • forge_validate_issue_url — GitLab URL regex, allowed hosts check
  • forge_parse_issue_url — extract GITLAB_HOST, REPO, REPO_ENCODED, ISSUE_NUMBER
  • forge_update_project_scores — attempt GitLab custom fields API via curl; if 403/404, log notice and return success (graceful degradation)
  • forge_post_comment — curl -X POST to /projects/${REPO_ENCODED}/issues/${ISSUE_NUMBER}/notes with PRIVATE-TOKEN header
  • Use _gitlab_api() helper pattern from gitlab-triage-ops.lib.sh

6. Pre-script — scripts/pre-prioritize.sh

• Source lib/prioritize-ops.lib.sh
• Replace hardcoded GITHUB_ISSUE_URL regex with forge_validate_issue_url "${ISSUE_URL}"
• Change notice line to use ISSUE_URL instead of GITHUB_ISSUE_URL
• Also create scripts/pre-prioritize.src.sh as the source file (current .sh has no .src.sh variant; the other agents use the .src.sh → bundled .sh pattern)

7. Post-script — scripts/post-prioritize.src.sh

• Source lib/prioritize-ops.lib.sh instead of lib/github-api-csma.lib.sh
• Replace GITHUB_ISSUE_URL with ISSUE_URL throughout
• Remove direct gh calls — replace with:
  • forge_validate_issue_url "${ISSUE_URL}"
  • forge_parse_issue_url "${ISSUE_URL}" (sets REPO, ISSUE_NUMBER as globals)
  • forge_update_project_scores "${REACH}" "${IMPACT}" "${CONFIDENCE}" "${EFFORT}" "${SCORE}"
  • forge_post_comment "${COMMENT}"
• Remove direct references to GH_TOKEN, ORG, PROJECT_NUMBER from the script body — those are consumed inside the GitHub ops lib
• Keep the shared logic: result file discovery, JSON validation, score extraction, RICE computation, comment body construction
• Rebuild bundled post-prioritize.sh via make script-build

8. Tests — scripts/post-prioritize-test.sh

• Update existing tests to set FULLSEND_FORGE=github and use ISSUE_URL
• Add GitLab test section:
  • Mock curl instead of gh
  • Assert no gh calls happen on GitLab forge
  • Test forge_validate_issue_url with GitLab URLs
  • Test forge_post_comment via curl
  • Test forge_update_project_scores graceful degradation (mock 403 response)

9. Docs — docs/prioritize.md

• "Scores a GitHub issue" → "Scores an issue"
• Mention GitLab support
• Note: on GitLab Premium/Ultimate, scores are written to custom fields; on Free tier, scores are posted as comments only

Verification

1. Run make script-build to regenerate bundled .sh files
2. Run scripts/post-prioritize-test.sh — all existing GitHub tests should pass with FULLSEND_FORGE=github
3. Run new GitLab test cases
4. Validate harness YAML with fullsend validate harness/prioritize.yaml (if available)
5. Manual smoke test: trigger prioritize on a GitLab issue and verify comment is posted

[fullsend-ai-triage]

🤖 Finished Triage · ✅ Success · Started 6:17 PM UTC · Completed 6:21 PM UTC

Commit: 7aa8587 · View workflow run →

[fullsend-ai-triage]

Triage Summary

Severity: medium · Category: feature

The triage agent multi-forge pattern is fully in place and provides complete templates for all 9 deliverables. Key reference files:

• scripts/lib/triage-ops.lib.sh — dispatcher pattern
• scripts/lib/gitlab-triage-ops.lib.sh — GitLab ops pattern (_gitlab_api() helper, forge_*() signatures)
• harness/triage.yaml — forge block structure
• policies/gitlab/triage.yaml — GitLab policy template
• env/gitlab/triage.env — GitLab env file template

Current state: The harness already has a partial forge.github block but no forge.gitlab block. The top-level policy: directive still exists and needs removal per the pattern.

Related multi-forge work: Code (#807/PR #813), review (#814/PR #815), and retro (#816/PR #817) conversions are open but are not prerequisites — each agent's conversion is independent, touching only its own file set.

GitLab score storage note: The graceful degradation approach (attempt custom fields API, log + succeed on 403/404) is consistent with how the platform handles tier-limited GitLab features.

# Proposed test: GitLab graceful degradation
# In post-prioritize-test.sh, GitLab section:
# Mock curl to return HTTP 403 for custom fields API
# Call forge_update_project_scores with valid RICE scores
# Assert: exit code 0, log contains "notice" about unavailable feature
# Assert: no gh binary calls occurred

---

Labels: This is a feature request for the prioritize agent. Apply 'feature', 'enhancement', and 'prioritize-agent' labels per repo conventions, plus 'ready-to-code' since the issue is fully specified with no open questions.

---

Next steps:

• /fs-code — agent creates a PR to implement this issue
• /fs-code <your instruction> — agent implements with your specific guidance

[ggallen]

/fs-code

[fullsend-ai-coder]

🤖 Finished Code · ❌ Failure · Started 7:41 PM UTC · Completed 7:53 PM UTC

Commit: 7aa8587 · View workflow run →

[fullsend-ai-coder]

⚠️ Post-code script failed — Pre-commit blocked (exit code 1)

The code agent completed, but the post-code script failed before finishing.

Workflow run: https://github.com/fullsend-ai/.fullsend/actions/runs/31904567853

Details:
export GITLAB_HOST="gitlab.com"
^---------^ SC2030 (info): Modification of GITLAB_HOST is local (to subshell caused by (..) group).

In scripts/post-prioritize-test.sh line 895:
    export FULLSEND_FORGE=gitlab
           ^------------^ SC2031 (info): FULLSEND_FORGE was modified in a subshell. That change might be lost.


In scripts/post-prioritize-test.sh line 896:
    export ISSUE_URL="${issue_url}"
           ^-------^ SC2031 (info): ISSUE_URL was modified in a subshell. That change might be lost.


In scripts/post-prioritize-test.sh line 897:
    export GITLAB_TOKEN="fake-gitlab-token"
           ^----------^ SC2031 (info): GITLAB_TOKEN [REDACTED] modified in a subshell. That change might be lost.


In scripts/post-prioritize-test.sh line 898:
    export GITLAB_HOST="gitlab.com"
           ^---------^ SC2031 (info): GITLAB_HOST was modified in a subshell. That change might be lost.

For more information:
  https://www.shellcheck.net/wiki/SC2030 -- Modification of FULLSEND_FORGE is...
  https://www.shellcheck.net/wiki/SC2031 -- FULLSEND_FORGE was modified in a ...

Lint GitHub Actions workflow files...................(no files to check)Skipped
pinact (SHA-pin check)...............................(no files to check)Skipped
lint agent doc references and structure..................................Passed

Please check the workflow logs for full details and retry with /fs-code if appropriate.

[ggallen]

/fs-code

Implement issue #818 following the multi-forge pattern from the triage agent (already merged).

Key reference files to study first:

harness/triage.yaml — forge block structure template
scripts/lib/triage-ops.lib.sh — dispatcher pattern
scripts/lib/gitlab-triage-ops.lib.sh — GitLab ops pattern (gitlab_api helper, forge* signatures)
policies/gitlab/triage.yaml — GitLab policy template
env/gitlab/triage.env — GitLab env file template

IMPORTANT — pre-commit will run shellcheck. The previous attempt failed on SC2030/SC2031 (subshell variable modification). In test files, do NOT use subshell groups (...) around exports. Use function-scoped setup or inline variable assignment instead. Study how the existing test files (e.g. scripts/post-triage-test.sh) handle per-test env setup to avoid this.

Run pre-commit run --all-files before committing to catch lint issues early.

Run make script-build after editing .src.sh files to regenerate bundled .sh files.

[fullsend-ai-coder]

🤖 Finished Code · ✅ Success · Started 8:03 PM UTC · Completed 8:18 PM UTC

Commit: 7aa8587 · View workflow run →