Skip to content

CHANGELOG.md

Latest commit

 

History

History
156 lines (125 loc) · 10.6 KB

CHANGELOG.md

File metadata and controls

156 lines (125 loc) · 10.6 KB
Raw

Changelog

All notable public changes to Team First Development are documented here.

Unreleased

3.5.0 - 2026-08-13

  • Add opt-in, session-only AUTO_REVIEW with one confirmed Autonomy Envelope, Orchestrator approval for in-envelope gates, bounded repair and evidence recovery, and no repeated user reply between approved steps.
  • Keep independent review as risk-scaled evidence, preserve exact external targets, and leave out-of-envelope, production, destructive, privileged, messaging, security, and structural Wiki changes human-gated.

3.4.1 - 2026-08-05

  • Remove all Superpowers Skill bindings from TFD documentation and example profiles; allow an explicit empty skills list when a member needs no extra working-method Skill.

3.4.0 - 2026-08-05

  • Make the Coordinator decompose a TEAM outcome into the minimum useful workstreams and map each one to exactly one accountable selected member.
  • Include the session-only workstream map, write scopes, dependencies, handoffs, and parallel or serial order in the exact confirmation proposal.
  • Allow only the Coordinator to activate members; prohibit member re-delegation and unconfirmed helper Agents.
  • Run only dependency-free, non-overlapping workstreams in parallel, stop dependent work after a BLOCKER, and integrate accepted outputs into one immutable candidate before review.

3.3.0 - 2026-08-01

  • Add one progressively loaded engineering contract for every source-code or architecture change, including SOLO work.
  • Require one concern owner, inward dependencies, fake or in-memory unit-test seams, one authoritative capability implementation, add/remove feature units, one canonical paradigm, current callers, and current-need minimalism.
  • Scale evidence depth with risk while retaining PASS, BLOCKER, and UNVERIFIED as the exact review verdicts.

3.2.0 - 2026-07-31

  • Add an exact per-member Runtime binding with codex_spawn, codex_thread, and codex_exec execution surfaces.
  • Require new or edited member profiles to declare runtime, while resolving legacy v3.1 profiles without it to codex_spawn without rewriting them.
  • Keep agency_role as optional frozen agency-agents provenance, independent from Runtime and model selection.
  • Use ~/.codex/models_cache.json only for account candidate discovery and require surface-specific eligibility evidence before activation.
  • Include exact Runtime, model, reasoning effort, evidence source, and user-visible behavior in the confirmation proposal.
  • Start the confirmed binding once, verify trusted Runtime metadata, preserve missing proof as UNVERIFIED, and prohibit automatic retry or fallback.

3.1.0 - 2026-07-31

  • Add a Git-native Markdown Team Wiki under each configured team, with immutable digest-addressed sources, maintained pages, exact proposal-confirm/apply knowledge writes, and no RAG or task control plane.
  • Add the standard-library Wiki Guard CLI for initialization, proposal validation/application, query records, bounded work activity, and strictly read-only integrity lint.
  • Record explicit user and team-member Wiki questions without a second confirmation; keep answers up to 1,000 Unicode code points in log.md and route longer answers to linked query detail files.
  • Merge bounded work_started, handoff, review, and delivery audit into the Wiki event chain while keeping knowledge promotion separately confirmed and retaining legacy receipts only before Wiki initialization.
  • Package the Team Wiki protocol under the using-tfd Skill so its on-demand relative reference resolves directly in a fresh Codex task.
  • Require one explicit confirmation of the exact team proposal before writing profiles or creating Agents; a generic team request or delegated choice does not confirm an unseen roster.
  • Allow one unchanged proposal to cover profile writes and immediate activation, while requiring renewed confirmation after any material proposal field changes.

3.0.0 - 2026-07-28

  • Breaking: replace the four-Skill v2 control plane with the suitability-first tfd:using-tfd and project-local YAML team profiles.
  • Configure reusable teams and named members without creating Agents; select at most one team and its minimum useful member subset for each requirement.
  • Require exact spawn_agent model choices and exact locally installed Skill names; stop on missing dependencies or initialization failure without substitution or fallback.
  • Remove persistent task state, state lifecycle, recovery, and management commands; retain only an optional final delivery receipt as a human-readable trace.
  • Preserve separate approval for installation, merge, push, tag, release, publication, and external actions.

2.1.3 - 2026-07-26

  • Derive aligned Task participants from the sealed TaskDocument, assign all and only that exact set through a prospective roster validation, and keep legacy Tasks owner-only.
  • Release exact participants atomically when a Task reaches DONE or CANCELLED, with an idempotent coordinator-release-task repair command for pre-v2.1.3 terminal Tasks.
  • Render current execution bindings from the authoritative roster while preserving Task ownership as separate history, and retain runtime preparation as a fail-closed dispatch prerequisite.

2.1.2 - 2026-07-25

  • Show users only a short human fingerprint for conversational approval while retaining the complete canonical digest internally for the exact machine/operator init-apply --approval-digest boundary; changed, stale, expired, mismatched, or ambiguous proposals require renewed confirmation.
  • Validate the user's outcome, shared success criteria, complete current phase, and each participant contribution when proposing task alignment before approval.
  • Disclose generated views as rebuildable rather than authoritative, and disclose nested Git repositories or worktrees while failing closed instead of treating them as ordinary candidate content.

2.1.1 - 2026-07-25

  • Add live create_thread model-allowlist evidence to TEAM scanning: pass observed exact IDs with repeated --create-thread-model flags, omit that input rather than reuse cached identifiers when metadata is unavailable, and prefer create_thread_eligible_models before the legacy eligible_models spawn_agent fallback.

2.1.0 - 2026-07-25

  • Assess every non-hidden live account model with sourced pros/cons and role-fit-first scoring, while keeping the optional Spark preference bonus bounded below role fit and execution hard gates.
  • Separate recommendations from execution: prefer verified create_thread capability, retain the exact-ID spawn_agent intersection as fallback, preserve unavailable backends as UNVERIFIED, and keep session evidence out of .tfd.
  • Preserve the complete recommendation catalog when the verified execution intersection is empty, and report execution separately as BLOCKER/no_verified_execution_backend.
  • Make the three product principles explicit: vision alignment, model-role fit, and context isolation. Persistent v1 records authoritative alignment, recipient TaskPackets are derived least-privilege projections, and a complete-participant gate blocks work until every required current-phase projection is valid.
  • Keep persistent new-v1, legacy compatibility, and session-only alignment routes separate: legacy records remain readable but UNVERIFIED, while session-only keeps alignment in active context, writes no .tfd/ path, and has no cross-session recovery.
  • Tighten reviewer and Coordinator authority boundaries: only Coordinator-owned, revision-safe state transitions can move authoritative task state; required reviewer/user PASS attestations bind the active reviewer or named user to the exact task revision and candidate digest.
  • Add an offline release preflight that rejects unsafe fixed release inputs, plus a cachebuster update flow that verifies the configured local marketplace source before a fresh task loads the rebuilt plugin cache.

2.0.0 - 2026-07-23

  • Breaking packaging change: replace the source standalone Skill with the installable tfd plugin. There is no compatibility shim; install the plugin from the repository marketplace and use tfd:using-tfd.
  • Split the workflow into a four-Skill architecture: the normal tfd:using-tfd entry coordinates the internal tfd:team, tfd:deliver, and tfd:workspace stages.
  • Keep the current primary Codex as the visible, non-roster Coordinator rather than a persistent Agent or task-team member.
  • Offer project-local .tfd/ persistence while preserving a complete session-only fallback when Workspace creation is declined.
  • Add authoritative Coordinator state, immutable Workspace events, and the deterministic, rebuildable .tfd/LOG.md timeline.
  • Preserve the local-only v1 boundary and separate approval for publishing, deployment, deletion, external messages, privileged commands, and external-system writes.

1.1.0 - 2026-07-23

  • Add the local-only TFD Workspace v1 CLI with proposal-first initialization and stable JSON envelopes.
  • Document authoritative versus generated state, Agent identity snapshots, capability-driven Skill review, default-deny whitelists, permission enforcement, immutable worklogs, archive/reactivation, and context-free recovery.
  • Document reviewed project_custom, BIND, INSTALL, and commit-pinned local Git checkout behavior.
  • Keep strict host invisibility honestly UNVERIFIED without concrete host evidence.
  • Verify one deterministic three-core-Agent/five-task local closed loop and 496 passing tests; named-user acceptance and out-of-scope deployment remain UNVERIFIED.
  • Retain digest-bound role sources separately from generated identity Markdown so initialized temporary experts can archive and reactivate safely.
  • Run every contract, catalog, and Workspace test through unittest discovery in CI.
  • Require immutable revision- and candidate-bound reviewer/named-user attestations for required team or user PASS, exposed through the local task-attest CLI.
  • Publish exclusive records from a fully flushed same-directory temporary file with atomic no-replace linking, directory fsync, and deterministic stale claim reconciliation.

1.0.0 - 2026-07-22

Initial public release.

  • Dynamically search and select local Agency Agents roles.
  • Support optional private custom-role catalogs.
  • Establish authority envelopes and team charters before execution.
  • Require progressive contract, vertical-slice, and frozen-candidate gates.
  • Bind review evidence to a reproducible Candidate manifest.
  • Normalize review and closeout to PASS, BLOCKER, and UNVERIFIED.
  • Keep automated, team, human/device, user, and deployment evidence independent.
  • Preserve upstream role sources as read-only during normal operation and tests.