apply-agentflow-comments

name	apply-agentflow-comments
description	Apply open AgentFlow architecture review comments, preserve resolved history, and resolve only comments visibly addressed by a revised proposal.

Apply AgentFlow Comments

Use this repo-local skill when the user asks to fix, address, resolve, or review AgentFlow architecture comments. This skill changes an architecture proposal, not implementation code.

Select The Proposal

1. Read AGENTS.md, relevant .agents/rules/, relevant .agents/memory/, and .agentflow/AGENTFLOW.md.
2. If the user named a proposal, feature, app, model, flow, node, edge, or comment, use that context to identify its proposal. Otherwise run:

python3 .agentflow/scripts/agentflow.py proposal list

3. Select the matching active proposal. Do not apply comments across unrelated proposals. If more than one active proposal plausibly matches and the request does not identify one, ask the user which review to change.
4. A proposal must be under_review or changes_requested to accept or resolve comments. Keep approved revisions immutable. When the user asks to change approved architecture, create a new proposal from the active architecture before recording or applying the new feedback.

Read The Right Comments

1. Import pending reviewer comments and read every Open comment on the selected proposal:

python3 .agentflow/scripts/agentflow.py comment list <proposal-id> --status open

This command imports comments saved from review.html into the canonical review before returning them, including pending reviewer edits and deletions. For every Open comment, read its target, current message, and proposal version. Do not leave an Open comment unexamined because it is outside the initially selected node or model. 2. Read Resolved comments only when they target an artifact or element that this revision will change, when the user refers to prior feedback, or when they are needed to avoid undoing an earlier decision:

python3 .agentflow/scripts/agentflow.py comment list <proposal-id>

Resolved comments are audit history, not pending work. Do not repeat their changes or alter their status just because they appear in the list. 3. If the user gives new feedback directly in chat, first record it as an anchored Open comment with comment add. Use the narrowest target that represents the feedback. See .agentflow/references/review-workflow.md for the command format and supported target kinds.

Apply The Feedback

1. Map each Open comment to a concrete architecture change in the proposal: an artifact, model field, flow node, edge, document section or entry, processing step, decision, job, event, or other represented element.
2. Read the approved baseline and proposal content needed to verify that the revised architecture will visibly address the comment. Do not rely on an intended code change or a prose promise.
3. If comments conflict, are ambiguous, or require a product decision, leave them Open and ask the user. Do not guess, silently narrow their scope, or resolve a blocked comment.
4. Create a complete revised proposal input, then revise and validate it:

python3 .agentflow/scripts/agentflow.py proposal revise <proposal-id> \
  --input /tmp/revised-proposal.json
python3 .agentflow/scripts/agentflow.py validate --proposal <proposal-id>

5. Resolve only the Open comments whose requested architecture change is present in the validated revision. Each resolution must name the actual change, not merely say that it was fixed:

python3 .agentflow/scripts/agentflow.py comment resolve \
  <proposal-id> <comment-id> \
  --resolution "Added the requested authorization decision before membership creation."

6. Re-list Open comments. Do not submit while any remain:

python3 .agentflow/scripts/agentflow.py comment list <proposal-id> --status open

7. Once the Open list is empty, submit and render the revised review:

python3 .agentflow/scripts/agentflow.py proposal submit <proposal-id>
make agentflow PROPOSAL=<proposal-id>

Status Rules

• open means the comment still requires an architecture response. Keep it Open when its change is incomplete, blocked, ambiguous, out of scope, or awaiting a user decision.
• resolved means the validated proposal visibly contains the requested change. Move a comment from Open to Resolved only with comment resolve after the revision is complete.
• The reviewer may edit or delete an Open comment before the agent applies it. Always run comment list first so the current canonical message and set of Open comments are authoritative.
• Resolved comments are preserved history and cannot be reopened through the CLI. If the user reverses or supersedes a resolved decision, add a new Open comment linked to the current target; do not edit the old comment or its resolution.
• Never edit .agentflow/comment-inbox/ or proposal review state by hand. Use .agentflow/scripts/agentflow.py so the reviewer and canonical history stay consistent.

Memory And Stop Point

Record only durable, reusable feedback in .agents/memory/architecture-feedback.md; do not copy every one-time comment into memory. After rendering the new review, report the comments resolved and any comments still Open or blocked. Stop before planning or implementation until the user explicitly approves the submitted architecture.