Skip to content

Per-agent model selection, in-session gate continuation, and web console agent-detail + operator mail - #81

Merged
stempeck merged 5 commits into
mainfrom
issues/80
Jul 7, 2026
Merged

Per-agent model selection, in-session gate continuation, and web console agent-detail + operator mail#81
stempeck merged 5 commits into
mainfrom
issues/80

Conversation

@stempeck

@stempeck stempeck commented Jul 7, 2026

Copy link
Copy Markdown
Owner

Summary

A single factory change set spanning several areas. Full detail, affected files, and
acceptance criteria per area are in #80.

  • Per-agent model selection — a models.json registry of named model/endpoint profiles with a
    per-agent assignment map and factory default. Per-launch --model override (sling/up) and
    dispatch-mapping pinning, resolved through one precedence chain that survives handoff, compact,
    and watchdog respawn. Endpoint credentials via file: references dereferenced at exec (never on
    the launch line or in the session env); loopback-vs-remote classification and a literal-credential
    warning. Session-env hygiene clears stale redirects on reused sessions, and legacy per-agent
    model/base_url/auth_token agents keep working. Quality/fidelity gate graders run insulated
    from the agent's redirection; the watchdog detects endpoint/gateway failures; af config models
    adds inspect/probe/preflight diagnostics.
  • In-session gate continuation — completing a gate step no longer ends the session; the agent
    continues via af prime (done.go, agent-shell generator, role templates).
  • Web console — agent-detail view — a per-agent page with a read-only, probe-first session
    snapshot (no cross-agent leak) and a gate banner.
  • Web console — operator mail — an injection-safe send path that messages an agent's mailbox as a
    reserved operator identity, backed by af mail send --from.
  • ultra-review forced re-review — an optional input that waives only the already-reviewed
    eligibility criterion.
  • New github-issue formula — a dispatchable agent that investigates a problem and produces an
    implementer-ready issue without prescribing the fix.
  • Supporting updates — manager verify-before-reporting rule, formula-create terminology
    refresh, quickstart/quickdocker default-model bump + cross-platform browser open + web auto-reveal,
    startup.json, and doc updates.

Testing

  • make sync-formulas / make sync-skills — source and installed copies in sync
  • make build — clean
  • make test — all root-module packages pass (non-integration)

Related

Closes #80

🤖 Generated with Claude Code

stempeck and others added 2 commits July 6, 2026 22:24
… --model override, dispatch pinning, file: secret handling, session-env hygiene, and gate-grader isolation. Switch gate completion to continue in-session via af prime. Add web console agent-detail view and operator mail, a github-issue formula, ultra-review forced re-review, and supporting workflow, installer, and doc updates.
@CLAassistant

Copy link
Copy Markdown

CLA assistant check
Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
1 out of 2 committers have signed the CLA.

✅ stempeck
❌ agentfactory


agentfactory seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
You have signed the CLA already but the status is still pending? Let us recheck it.

@stempeck
stempeck merged commit 50fe9c5 into main Jul 7, 2026
5 of 6 checks passed
@stempeck
stempeck deleted the issues/80 branch July 7, 2026 02:32
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

2 participants