Goal at the End of the Previous PR

Post #123 landed the qa-merge-gate-authority stack: synlynk pr check computes a real block/pass verdict from CI + sentinel state, and apply_qa_gate_branch_protection.sh exists (dry-run verified, not yet applied for real). Three things were explicitly deferred: applying branch protection for real, landing #1079's original design spec doc, and the follow-up work each of those specs called out as "not built yet" — including a reserved-but-unbuilt qa_gate_mode value (merge-restricted-classes) and a way to verify that a merged PR actually fulfilled the spec/plan it claimed to implement, not just that its own tests passed.

No Strategic Shift

Both plans executed this PR were written and approved before #123 merged (docs/superpowers/plans/2026-08-22-qa-completion-tracker.md off spec #1091, docs/superpowers/plans/2026-08-22-qa-merge-restricted-classes.md off spec #1079 §5). Nikhil approved running both in parallel via subagent-driven-development once #123 was stable. No goalpost moved — this is the two follow-ups #123 named, built.

What Shipped, Technically

PR #1100 — Completion Tracker (qa-completion-tracker plan, 4 tasks):

• synlynk/completion_tracker.py: parse_spec_reference() extracts a spec/plan path or issue reference out of a PR body; compute_completion_verdict() + _load_reference_content() produce a fulfilled/partial/diverged verdict with rationale by comparing the referenced spec against the actual PR diff.
• synlynk/events.py: _scan_pr_completion() + _existing_spec_verified_pr_numbers() wired into scan_local_events() — mirrors the existing review_submitted scan pattern from the GOVERNS event-contract extension, finding pr_merged events with no matching spec_verified event yet and emitting one once a verdict is computed.
• synlynk/viz.py: new Vizor panel on the existing BS-21 stage-progress bars — _load_spec_verifications() loader, spec_verifications key in _base_data(), renderVerified() / toggleVerifiedPanel() / renderVerifiedEntry() JS, plus a summary card and panel in the HTML — so merged-but-not-yet-verified PRs are now visually distinct from merged-and-verified ones, closing the gap #123 flagged: a dispatched agent's own tests passing was never the same claim as "this PR actually did what the spec said."
• Tests: tests/test_completion_tracker.py, tests/test_completion_tracker_vizor.py, extensions to tests/test_events.py.

PR #1101 — Merge-Restricted-Classes Gate Mode (qa-merge-restricted-classes plan, 4 tasks):

• synlynk/merge_class.py: is_docs_only_change() — a file-pattern matcher that classifies a PR's changed files as docs-only or not.
• synlynk/qa_gate.py: _qa_gate_mode() reads qa_gate_mode from .synlynk/config.json (default "block-only"); _gh_pr_changed_files() wraps gh pr diff <n> --name-only. A real bug surfaced and was self-corrected mid-task here: the first commit (353e38e) read synlynk/config.json instead of .synlynk/config.json; a follow-up commit (a37e3e8) fixed it before the PR was finalized — caught by the implementer's own review pass, then independently re-verified in this session's spec-compliance review.
• synlynk/db.py: wired the fast path into cmd_pr_check() — when qa_gate_mode == "merge-restricted-classes" and the PR is docs-only, qa merges it directly rather than routing through the full block-only gate. Task 3 also added module-level wrapper functions (_qa_gate_mode, _gh_pr_changed_files, _is_github_remote, _current_pr_number, _extract_pr_review_cycles, _apply_review_cycle_multiplier) re-exporting the real implementations from qa_gate.py/pr_multiplier.py/sentinel.py — necessary, not scope creep: the plan's own test fixtures patch these exact names as synlynk.db.* module attributes, which requires them to exist at module scope.
• Task 4: since .synlynk/config.json is gitignored, this task landed as a spec-doc note (in the gate-authority design doc) rather than a code default — correctly scoped to what's actually committable.

Both PRs were reviewed directly by Claude (spec-compliance first, then code quality — no sub-delegated review) against their authoritative plan text, comparing diffs section-by-section rather than trusting the implementer's own summary. Zero blocking findings on either. Both merged via the #423 COMMENT-review-with-checklist fallback (gh pr review --approve fails under the shared bot identity), then gh pr merge --squash.

Process Note: Dispatch Stacking Produced 8+ PRs for 2 Plans

Both plans stacked 4 tasks each via synlynk dispatch <agent> --base <branch>. Every stacked task opened its own PR against main rather than against the parent task's branch — not a one-off, a --base behavior. The two 4-task plans produced 8+ PRs and 16 worktrees/branches total, requiring a full Worktree Hygiene Protocol sweep after both terminal PRs (#1100, #1101) merged: cross-referencing every branch against gh pr list --state all and, since squash merges break literal ancestry, validating no-PR branches against their pre-squash terminal branch tips instead of origin/main directly. All 16 worktrees and local branches removed, 12 remote branches explicitly deleted (4 had already been auto-pruned by GitHub on merge/close). This is now recorded as a standing process note for future stacked dispatch sessions — worth a synlynk issue to make --base target the parent branch as PR base instead of always opening against main.

Brainstorm Visuals Used

None — both plans executed directly off their already-approved spec/plan documents; no new brainstorming was needed.

What This Achieved on the Path to Autonomy

The completion tracker closes a trust gap the qa-gate stack (#123) opened but didn't close: a green qa-gate proves CI passed, not that the PR did what it claimed. Vizor now distinguishes merged from verified, giving a human (or eventually another agent) a place to see that distinction without re-reading every diff. The merge-restricted- classes mode is the first concrete, narrow answer to #1079 §5's deferred question — "what PR class is safe to let qa merge unattended" — landing exactly the example the original spec named (docs-only) rather than the broader, harder case (dependency bumps, CI config) that still needs its own design pass.

Strategic Note: The Goal at the End of This PR

Both #1091 and #1079's design-doc PRs (still open) now have their concrete follow-up plans fully implemented, reviewed, and merged. What remains, unchanged from #123's own deferred list: actually applying branch protection for real (still a deliberate human-confirmed step, not yet scheduled), and landing #1079/#1091 themselves as merged design docs rather than leaving the implementation ahead of the spec. The new goalpost: extend qa_gate_mode == merge-restricted-classes to the harder PR classes #1079 §5 named but deferred (dependency bumps, CI config), now that the docs-only case has a working, reviewed reference implementation to build from.