chore: bump Workforce pins to 4.1.43 so the relayflows dep resolves honestly

[khaliqgant]

Follow-up to #112, which Codex flagged as P1 — correctly.

The problem

#112 added @relayflows/core@^1.0.6, which declares:

@agentworkforce/persona-kit      ^4.1.39
@agentworkforce/persona-registry ^4.1.39

while this repo's global override pinned persona-kit to 4.1.23. npm overrides suppress that incompatibility, so npm install succeeds while core/registry load against a persona-kit below their declared minimum — nothing fails until agent startup.

The change

Every @agentworkforce/* pin and the agentworkforce CLI: 4.1.23 → 4.1.43, in both dependencies and overrides, so no package is left resolving behind an override.

$ npm ls @relayflows/core @agentworkforce/persona-kit --depth=1
└─┬ @relayflows/core@1.0.6
  ├── @agentworkforce/persona-kit@4.1.43 deduped
  └── @agentworkforce/persona-registry@4.1.43

Ranges satisfied rather than overridden.

The bump surfaced a real typing break

Which is rather the point of doing it. At 4.1.43, ctx.persona.id is unknown:

WorkforcePersonaContext extends Omit<PersonaSpec, 'inputs'>
PersonaSpec { …; [key: string]: unknown }   // consumer-defined fields

Omit over an index signature collapses every declared field — id, description, intent — to unknown. aliasVariants in linear/agent.ts now takes unknown and narrows, which is also strictly safer than the old string | undefined signature that would have thrown on .trim() had a non-string ever arrived.

That collapse looks like an upstream typing defect worth fixing in runtime/persona-kit rather than worked around at every call site — any handler reading ctx.persona.<field> hits it. This PR only unblocks the one site here.

Verified

• tsc --noEmit — 0 errors
• npm test — 250/250
• npm run compile — all 8 personas compile

Two caveats, both environmental

Neither is attributable to this change, and I'd rather flag them than imply a clean sweep:

1. npm test initially failed 2/250 with ENOENT workforce/packages/harness-kit/package.json. harness-kit was deleted upstream (workforce 451a5305) and my checkout retained an orphaned untracked dist/. workforce-package-proof.mjs enumerates packages/* and reads each package.json, so the orphan breaks it. Moving the two orphan dirs (harness-kit, relay-helpers) aside gave 250/250; restored afterwards. Suggest the script skip directories without a package.json so a stale checkout can't fail the suite.
2. acceptance:composable-runtime-closure exits 1 after ~20s with no output. The globally installed agentworkforce CLI here is 4.1.22 against repo-local 4.1.43, and the harness invokes it. I couldn't attribute the failure to this change, so I'm leaving it for CI rather than claiming a pass I hadn't established.

🤖 Generated with Claude Code

---

Summary by cubic

Aligns all @agentworkforce/* pins and the agentworkforce CLI from 4.1.23 to 4.1.43 so @relayflows/core@^1.0.6 resolves against supported persona-kit/persona-registry versions instead of being masked by overrides. Updates linear/agent.ts to accept unknown for persona fields introduced by persona-kit@4.1.43; previously aliasVariants assumed a string and could throw, now it narrows safely.

Review notes

• Dependencies and overrides both pin @agentworkforce/* to 4.1.43; npm ls shows @relayflows/core@1.0.6 deduped to @agentworkforce/persona-kit@4.1.43 and @agentworkforce/persona-registry@4.1.43.
• aliasVariants(value: unknown) adds a typeof check; behavior is unchanged for strings and returns [] for non-strings.
• If running acceptance locally, upgrade your global agentworkforce to 4.1.43.

^{Written for commit 824bd20. Summary will update on new commits.}

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.

[cubic-dev-ai]

No issues found across 3 files

<sub>Tip: cubic can generate docs of your entire codebase and keep them up to date. Try it here.

Re-trigger cubic</sub>

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 824bd20187

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Align the Linear adapter with the bumped helpers

When the bumped runtime or delivery package loads its Relay helpers, the new lock installs @relayfile/relay-helpers@0.4.11, which declares @relayfile/adapter-linear@^0.4.11 (package-lock.json lines 393-401 and 516-524), but the existing global override still forces the only adapter copy to 0.3.11. This recreates the same override-suppressed unsupported dependency combination that this bump is intended to eliminate and can break Linear-backed agents at startup; remove or update the adapter override (and regenerate the lock) as part of the Workforce bump.

Useful? React with 👍 / 👎.

[khaliqgant]

Fixed in the latest commit — good catch, and it's the same class of bug this PR exists to remove. I eliminated one override-suppressed conflict and left an identical one immediately beside it.

It was in fact worse than reported: our direct @relayfile/relay-helpers@0.4.6 already declared adapter-linear ^0.4.5, so the 0.3.11 override sat below both ranges before this bump. The bump widened the gap rather than creating it.

@relayfile/adapter-linear  override   0.3.11 -> 0.4.12
@relayfile/relay-helpers   dependency 0.4.6  -> 0.4.11

$ npm ls @relayfile/relay-helpers @relayfile/adapter-linear --depth=2
├─┬ @agentworkforce/runtime@4.1.43 overridden
│ ├─┬ @agent-relay/events@6.3.6
│ │ └── @relayfile/adapter-linear@0.4.12 deduped
│ └── @relayfile/relay-helpers@0.4.11 deduped
└─┬ @relayfile/relay-helpers@0.4.11
  └── @relayfile/adapter-linear@0.4.12 overridden

No range is now satisfied only by an override. The 0.3.11 pin arrived incidentally with the eval-harness PR (#58) and carried no recorded reason to hold that version.

Re-verified: tsc --noEmit 0 errors, npm test 250/250, npm run compile all 18 personas.

Two corrections to my PR description while I'm here:

• it says "all 8 personas compile" — it's 18; I read a truncated tail and undercounted.
• it says the acceptance-harness failure is left "for CI to adjudicate". This repo has no test CI (only pullfrog.yml, a review bot), so nothing was going to run it. That was an empty deferral on my part. The harness exits 1 silently after ~20s with a globally-installed agentworkforce CLI at 4.1.22 against repo-local 4.1.43; it invokes that CLI, which is the most likely cause, but I have not proven it and it remains genuinely unverified rather than deferred.