Add Slack post-to-X workflow

[pratikbuilds]

Summary

• Add a Slack post-to-X starter using the pinned Corbits Tag checkout introduced by Use Corbits Tag for the Interchange Slack agent #14
• Reuse that single shared checkout instead of registering another vendor gitlink
• Draft with one inference step, then validate and publish through deterministic Interchange actions
• Require correlated Slack approval before the capability-gated X effect
• Default to dry-run mode and avoid retries when an X write may have an unknown outcome

Stack

Depends on #15 and targets its head branch,
agent/use-corbits-tag-for-slack-workflow.

This branch is authored directly on top of PR #15's head commit,
7bd009c1aa58737e490c5ab00f24ee45ee0fa66c.

Verification

• bun install --frozen-lockfile
• bun run typecheck
• bun run start --help
• Confirmed the starter has no second Corbits Tag gitlink, bridge, or twitter-text dependency

No live Slack or X request was made.

[TheGreatAxios]

Self-review (agent)

Verdict: Request changes

What this PR does

Adds starter/slack-post-to-x: one inference draft, deterministic validate/publish, Slack approval before a capability-gated X write. Defaults to dry-run (X_LIVE=1 required for live). Reuses the shared Corbits Tag workspace from the stack.

Also ships a stalled-session hardening for slack-approval-flow (see below) and branding/license touch-ups on sibling starters.

X write safety — looks good

• Dry-run unless X_LIVE is exactly "1"; missing OAuth + live fails at config resolve
• No auto-retry when X write outcome is unknown (network/body/id missing)
• Publish uses workflow steps.policy.output, not editable card text
• Effect + authorize: x:post capability, deny-by-default
• Approval correlation: approvalId + message id + unset decision; decision set before signal

Findings that block merge

Severity	Finding
major	Commit cef0c3f added approval-flow session tests (failed draft, timeout, stale action); 0a2d4f6 deletes them. Production fix ships with no regression coverage — please restore.
major	New post-to-x starter has zero tests on the highest-risk write path (dry-run, unknown outcome, validation, authorize deny, session decide).
major (scope)	Branding rename (interchange → corbits*) and MIT → LGPL-2.1-only on sibling starters are not needed for post-to-X. Split into their own PR or drop from this stack.

Other findings

Severity	Finding
minor	post-to-x did not inherit the stalled-session hardening applied to approval-flow (draft timeout, failed-draft observer, ephemeral on stale approve). The write-capable starter is the more important place for it.
minor	bun.lock is committed while listed in .gitignore (confusing for contributors).
history	Revert/reapply/merge thrash + triple branding commits — please squash before merge.

Stalled-session hardening (what landed on approval-flow)

Before this PR, a draft that never finished or failed quietly could leave the thread’s active map entry stuck, and stale Approve/Reject buttons did nothing with no user feedback. This PR adds:

1. 60s draft timeout — fails the wait instead of hanging forever
2. Draft failure observer — draft step errors surface instead of racing only on success
3. Ephemeral on unavailable actions — “This approval is no longer active.”
4. Clearer map ownership (clear(pending)) so a failed run frees the thread

That production change is good; the deleted tests are what make it a request-changes.

Residual risks (documented, OK for a demo)

• Process-local effect ledger + active map: after restart old buttons are ignored, but a new live run of the same text after an unknown X outcome can still double-post — operator must inspect X first
• Any channel member who can click Approve can publish live (no requester/role check)

Checks

• Full incremental review of 27 files (+1890/−48)
• Read of x-client.ts, actions.ts, session.ts, workflow.ts, post.ts + approval-flow session delta
• Commit graph verified for test add/delete

Please restore the approval-flow tests, add minimal post-to-x coverage (or justify omission), and move branding/LGPL out of this PR — then happy to re-review.

[TheGreatAxios]

Review · Approve

Adds starter/slack-post-to-x: one inference draft, deterministic validate/publish, Slack approval before a capability-gated X write (dry-run default). Also hardens stalled approval sessions on slack-approval-flow with tests, and ports the same session hardening to the write-capable starter.

Findings

None that block merge.

Prior request-changes (resolved)

• Approval-flow session tests restored (session.test.ts); bun test → 3 pass
• Write-path coverage added in x-client.test.ts (dry-run default, live OAuth gate, unknown outcome not retried); bun test → 4 pass
• Branding/LGPL scope creep gone from this fork head vs base
• Post-to-x now has draft timeout, failed-draft surface, ephemeral on stale actions
• bun.lock no longer gitignored while committed

Checks

• Diff vs agent/use-corbits-tag-for-slack-workflow (21 files, +2114/−34); no unexpected binaries
• 4 commits, plain-English subjects, bodies empty, none over 72 chars
• bun install --frozen-lockfile, bun run typecheck, bun run start --help in slack-post-to-x
• bun test + typecheck in slack-approval-flow

Residual (documented / acceptable for a starter)

• Process-local active map and effect ledger: restart invalidates old buttons; operator must inspect X after an unknown live outcome before re-running
• Any channel member who can click Approve can publish when live
• Validation failures surface under “Workflow failed” rather than the unused “Validation failed” branch in session.ts (validationError is set but the reject path never reaches that card)
• Post-to-x session correlation path is untested (write safety is covered in x-client.test.ts; approval-flow has the session suite)