git:20260904.2014e73 to git:20260905.d71dd04

35 added, 16 removed. Audit A to A.

---

name: resolve

- description: 'Use when addressing review feedback or analyzing whether a comment is valid. Not for GitHub PR review threads: use resolve-pr-feedback.'

+ description: 'Use when addressing review feedback: analyze mode classifies comment validity, reception mode applies accepted items with tests and drafts pushback. Not for GitHub PR threads: use resolve-pr-feedback.'

---

# Resolve

## Contract

| Field | Bound contract |

|---|---|

| Trigger | addressing review feedback or analyzing whether a comment is valid; for GitHub PR threads use resolve-pr-feedback |

- | Authority | Read-only. No file, VCS, credential, paid, published, deployed, or remote mutation. |

- | Side effect | produces analysis and proposed solutions in chat; no code edits and no remote calls |

- | Done | each comment classified VALID ISSUE / NOT AN ISSUE / NEEDS CLARIFICATION with evidence, and a recommended solution for VALID ISSUE |

+ | Authority | Reversible local. Mode analyze is read-only. Mode reception writes only local source files, test files, and pushback drafts; rollback is version control or undo. No remote mutation. |

+ | Side effect | analyze: classification and proposed solutions in chat, no code edits. reception: local edits, tests, and a draft file; no GitHub reply, CI trigger, or issue |

+ | Done | every comment is classified; analyze: VALID ISSUE / NOT AN ISSUE / NEEDS CLARIFICATION with evidence and a recommended solution for each VALID ISSUE. reception: each item is clarified, implemented locally with its own test, or answered by a drafted pushback |

## Inputs

- The user supplies review comments: the text of each comment and any associated code context. The user also supplies the location in the codebase under review, or the model reads it directly using only read-only tools.

+ - Mode: `analyze` or `reception`. Required. Default `analyze` when the user asks only whether comments are valid.

+ - Feedback: the raw review comment(s), PR thread, or inline suggestion. Required.

+ - Code under review: the relevant source files or their location. Required.

+ - PR or issue thread: the full review context. Optional; use when available.

## Procedure

- 1. Parse every review comment into an isolated unit. Stop on an empty or unparseable comment list; return "No comments supplied" without hallucinating any.

+ 1. Parse every review comment into an isolated unit. Do not combine, rephrase, or infer unstated items. Done when: each comment stands as a separate item, or the run stops with "No comments supplied" on an empty or unparseable list.

- 2. For each comment, explore the referenced code location using only read-only tools: `grep`, `read`, `glob`, or shell equivalents (`eza`, `rg`, `fd`, `head`, `tail`). Never invoke `bash` with write operations (`mkdir`, `touch`, `rm`, `cp`, `mv`, `git add`, `git commit`, `npm install`, redirect operators `>`, `>>`, or heredocs that create files). Do not use `ast-grep` or any tool with write, rename, delete, or staging arguments.

+ 2. Read the referenced code. Mode analyze: use only read-only tools (`grep`, `read`, `glob`, or shell equivalents such as `eza`, `rg`, `fd`, `head`, `tail`); never invoke `bash` write operations (`mkdir`, `touch`, `rm`, `cp`, `mv`, `git add`, `git commit`, `npm install`, redirects `>`, `>>`, or file-creating heredocs) or any tool with write, rename, delete, or staging arguments. Mode reception: read freely; writes are confined to steps 5 and 6. Done when: the code context for each item is read, or the item is marked unreadable.

- 3. Classify each comment:

- - VALID ISSUE: the concern is confirmed by the code. Propose three distinct solutions with trade-offs and name the recommended one with justification.

- - NOT AN ISSUE: the concern is not supported by the code or contradicts project patterns. Supply evidence from the code.

- - NEEDS CLARIFICATION: the comment cannot be assessed without additional context. State exactly what is missing.

+ 3. Clarify each item to a concrete statement. Mode reception: resolve the precise intent from context or by asking the user; stop on any item that cannot be resolved to an implementable statement. Done when: every item is a concrete, assessable statement.

- 4. Stop before proposing solutions for any comment classified NOT AN ISSUE or NEEDS CLARIFICATION.

+ 4. Classify each item. Done when: every item carries exactly one classification with evidence.

+ - Mode analyze:

+ - VALID ISSUE: the concern is confirmed by the code. Propose three distinct solutions with trade-offs and name the recommended one with justification.

+ - NOT AN ISSUE: the concern is not supported by the code or contradicts project patterns. Supply evidence from the code.

+ - NEEDS CLARIFICATION: the comment cannot be assessed without additional context. State exactly what is missing.

+ - Stop before proposing solutions for any comment classified NOT AN ISSUE or NEEDS CLARIFICATION.

+ - Mode reception:

+ - Accepted: the suggestion aligns with the codebase and the user's intent.

+ - Questionable: the suggestion is technically incorrect, out of scope, or conflicts with design intent.

+ 5. Mode reception: apply accepted items one at a time. Apply the change to the relevant source file, write or update a test that covers the change, and verify the test passes before moving to the next item. Done when: each accepted item is implemented with a passing test.

+ 6. Mode reception: draft factual pushback for questionable items in a local draft file. State the specific reason the suggestion is not accepted, such as a design conflict, incorrect assumption, or scope mismatch. Do not post it. Done when: each questionable item has a drafted reply.

+ 7. Stop. Do not post any GitHub reply, open any issue, trigger any CI pipeline, or otherwise mutate any remote system. Done when: no remote state has changed.

## Failure and recovery

- Empty input: return "No comments supplied" and stop.

- - Inaccessible code: if the referenced location cannot be read, classify the comment NEEDS CLARIFICATION with the specific access failure as the missing context.

- - Unresolvable ambiguity: classify NEEDS CLARIFICATION; do not guess intent or invent a fix.

+ - Inaccessible code: if the referenced location cannot be read, classify the comment NEEDS CLARIFICATION in analyze mode, or report it unclarified in reception mode, with the specific access failure as the missing context.

+ - Unresolvable ambiguity: analyze: classify NEEDS CLARIFICATION; do not guess intent or invent a fix. reception: stop and report which item cannot be clarified to a concrete statement; do not implement it.

- Partial-result rule: if one comment cannot be processed, continue with the remaining comments and report the failure for the unprocessed one. Do not claim the whole batch succeeded.

- - Non-mutation rule: no edit, write, commit, push, deployment, credential use, or remote call is ever performed.

+ - User rejects an accepted item (reception): skip that item; continue with the remaining items.

+ - Implementation blocked (reception): report the specific technical obstacle; do not widen scope.

+ - Test fails (reception): fix the implementation, not the test. If the test is wrong, report it and stop.

+ - Remote mutation attempted: refuse; log the blocked action.

+ - Non-converged result: any unresolved failure produces a report listing unresolved items and blocked actions. The done predicate does not hold.

## Output

- Per comment, in order: Comment, Status (VALID ISSUE | NOT AN ISSUE | NEEDS CLARIFICATION), then the status-specific block, for VALID ISSUE three numbered Solutions with trade-offs and the Recommended pick; for NOT AN ISSUE the Reason and Evidence; for NEEDS CLARIFICATION the exact Missing information. Stop after the last comment; emit no summary or overall-PR content.

+ - Mode analyze: per comment, in order: Comment, Status (VALID ISSUE | NOT AN ISSUE | NEEDS CLARIFICATION), then the status-specific block: for VALID ISSUE three numbered Solutions with trade-offs and the Recommended pick; for NOT AN ISSUE the Reason and Evidence; for NEEDS CLARIFICATION the exact Missing information. Stop after the last comment; emit no summary or overall-PR content.

+ - Mode reception: a local report listing every feedback item, its classification (accepted or questionable), the actions taken for each accepted item, and the location of any drafted pushback. No remote state is changed.