Repository navigation
Deferred follow-up: Human-initiated review remediation worker #1735
Description
Related epic: #1729
Depends on: #1734 enforcement semantics and operational evidence from #2096
Disposition
Valid deferred follow-up — optional, not duplicate, and not part of the current review-gate merge train.
PR #2096 implements enforcement, exact-head verdict gating, logical reviewer independence, and an auditable human override. It does not provide branch remediation: its override can clear an actual rejection for an exact SHA, but it does not edit the rejected PR branch. Its changed-file set contains workflows/squad-review.md and workflows/squad-implement-worker.md, but no workflows/squad-review-remediate.md or equivalent edit-capable workflow.
The remaining behavior is therefore specific and unmet: after REQUEST_CHANGES, an authorized human should be able to explicitly request a separate, edit-capable agent to revise the PR head branch without using the rejected author or allowing the remediation run to clear its own verdict.
This issue is not a gate for #2096 hosted validation, #1734 completion, or #1729 closure. Hosted validation should prove the enforcement/override behavior implemented by #2096; it must not claim that automated remediation exists. Revisit this follow-up only if operational evidence shows that manual author revision is insufficient.
Goal
Provide an optional way to act on a REQUEST_CHANGES verdict without a human hand-editing the branch, while preserving reviewer independence.
Proposed design
A separate workflow, workflows/squad-review-remediate.md, holds the edit: authority that squad-review.md deliberately lacks. It is dispatched only by an explicit authorized human command (/squad review fix), never automatically by the reviewer.
Acceptance criteria
- New workflow with
edit:authority scoped to the PR head branch - Triggered only by an explicit authorized human slash command, never by the review workflow
- Same file allowlist and
protected-filesposture assquad-implement-worker.md - Re-review is required after remediation; the remediation run cannot clear its own verdict
- Rejected-author lockout is honored: the remediating agent must not be the logical agent that authored the rejected diff
- Tests prove the human-trigger-only constraint and reject reviewer/self-remediation paths
- Documentation distinguishes remediation from the exact-SHA human override delivered by fix(gh-aw): enforce independent Squad review and gate epic relay #2096
Why deferred
An automatic review-then-fix loop collapses reviewer and author roles. The current safe path is author/human revision followed by re-review, plus the narrowly scoped human override in #2096 when the reviewer is wrong. Do not add remediation to the current hosted-validation scope or enable it without a separate authority and threat-boundary review.
Activity
bradygaster commentedon Aug 23, 2026
Procedures triage brief — Phase 3 remediation worker
Owner & rationale
Owner: EECOM (squad:eecom). This is an authoring workflow with �dit: authority on PR branches, so it belongs with core workflow/runtime mechanics. FIDO should review the independence lockout.
Files in scope
Verified existing paths:
New target path to create during implementation: workflows\squad-review-remediate.md inside the verified workflows\ directory.
Prompt-budget impact
No expected workflows\squad.md growth: +0 KB. Keep remediation separate; adding fix-loop instructions to the router would spend finite prompt budget and blur the authority boundary.
Dependencies
Verification bar
A test must fail if remediation can be triggered by the review workflow, by a non-human actor, or by the same agent that authored the rejected diff. A remediation run must not be able to clear its own reviewer verdict.
Out of scope
Automatic review-then-fix loops, required-check setup, and documentation.
Session-stall guard
Before starting: this repo has phantom CRLF diffs (#1793). If
git statusshows files modified that you did not touch, rungit diff --ignore-cr-at-eol. If that output is empty, the diff is line-ending churn, not real edits — rungit add --renormalize .and continue. Do not halt and ask for confirmation; halting on phantom diffs is the documented cause of stuck sessions in this repo.bradygaster commentedon Aug 30, 2026
Dependency note (Ralph triage): This phase (optional human-initiated remediation worker) is dependent on #1734. The lockout behavior design requires #1734 verdict/override semantics to be settled first. No action until #1734 unblocks.
github-actions commentedon Sep 10, 2026
🔄 Ralph — Auto-Triage
Assigned to: Flight
Reason: No module, routing, or role keyword match — routed to Lead/Architect
Source: lead-fallback
12 remaining items
github-actions commentedon Sep 13, 2026
🔄 Ralph — Auto-Triage
Assigned to: architect
Reason: No module, routing, or role keyword match — routed to Lead/Architect
Source: lead-fallback
github-actions commentedon Sep 24, 2026
📋 Assigned to agentic-workflows-dev (Agentic Workflows Engineer)
Issue: #1735 — Phase 3: Optional human-initiated review remediation worker
agentic-workflows-dev will pick this up in the next Copilot session.
[-]Phase 3: Optional human-initiated review remediation worker[/-][+]Deferred follow-up: Human-initiated review remediation worker[/+]bradygaster commentedon Sep 28, 2026
Disposition: keep as a deferred, optional follow-up. It is not a duplicate of #1734/#2096: #2096 supplies exact-head enforcement, relay gating, distinct logical identities, and a human override, but its file set contains no edit-capable remediation workflow and the override does not revise a PR branch. The unmet behavior is an explicit human-triggered, separately authored branch fix followed by independent re-review. I removed this issue from the M4 milestone and marked it deferred/shelved/backlog so it does not gate #2096 hosted validation, #1734 completion, or #1729 closure. Hosted evidence must not claim remediation automation exists.
bradygaster commentedon Sep 29, 2026
Post-#2105 reconciliation
No edit-capable human-triggered remediation worker was added by the merged gh-aw stack. The current design intentionally provides enforcement and exact-SHA override, not automated remediation. This remains deferred/shelved and should be revisited only if hosted operational evidence justifies the additional authority.
bradygaster commentedon Nov 5, 2026
Closing during 2.0 board triage (owner decision). The idea is evaluated, with prior art and an opinion, in the roll-up issue #2390. Original text remains here for reference.