Update answer-questions-first with workflow awareness, pr-land reply step, author behavioral dependency checks; add lint-warnings.sh to README

Author: Jon-edge

.cursor/README.md

@@ -153,6 +153,7 @@ These scripts run sequentially. Each handles one phase of the landing workflow:
 | Script | What it does |
 |--------|-------------|
 | [`lint-commit.sh`](.cursor/commands/lint-commit.sh) | ESLint `--fix` before commit, auto-runs `update-eslint-warnings` when available |
+| [`lint-warnings.sh`](.cursor/commands/lint-warnings.sh) | Update `eslint-warnings.mdc` knowledge base from current lint output |
 | [`upgrade-dep.sh`](.cursor/commands/upgrade-dep.sh) | Upgrade a dependency in the GUI repo |
 
 ---
@@ -182,7 +183,7 @@ These scripts run sequentially. Each handles one phase of the landing workflow:
 | [`review-standards.mdc`](.cursor/rules/review-standards.mdc) | Loaded by `/pr-review` command | ~50 review-specific diagnostic rules extracted from PR history |
 | [`load-standards-by-filetype.mdc`](.cursor/rules/load-standards-by-filetype.mdc) | Always applied | Auto-loads language-specific standards before editing |
 | [`fix-workflow-first.mdc`](.cursor/rules/fix-workflow-first.mdc) | Always applied | Fix command/skill definitions before patching downstream symptoms |
-| [`answer-questions-first.mdc`](.cursor/rules/answer-questions-first.mdc) | Always applied | Detect `?` in user messages → answer before acting |
+| [`answer-questions-first.mdc`](.cursor/rules/answer-questions-first.mdc) | Always applied | Detect `?` in user messages → answer before acting; loads active command context to evaluate workflow gaps |
 | [`no-format-lint.mdc`](.cursor/rules/no-format-lint.mdc) | Always applied | Don't manually fix formatting — auto-format on agent finish handles it |
 | [`eslint-warnings.mdc`](.cursor/rules/eslint-warnings.mdc) | `.ts`/`.tsx` files | ESLint warning handling patterns |
 
@@ -194,7 +195,7 @@ These scripts run sequentially. Each handles one phase of the landing workflow:
 
 | Skill | Purpose |
 |-------|---------|
-| [`author/SKILL.md`](.cursor/skills/author/SKILL.md) | Meta-skill for creating/maintaining commands and skills. Enforces XML format, `scripts-over-reasoning`, `gh-cli-over-curl`, `minimize-context`, companion script naming conventions, and `small-model-conventions` for commands targeting faster models. |
+| [`author/SKILL.md`](.cursor/skills/author/SKILL.md) | Meta-skill for creating/maintaining commands and skills. Enforces XML format, `scripts-over-reasoning`, `gh-cli-over-curl`, `minimize-context`, companion script naming conventions, `small-model-conventions`, and behavioral dependency checks during revision. |
 
 ---
 

.cursor/commands/pr-land.md

@@ -77,9 +77,10 @@ Items previously marked with `<!-- addressed:review:ID -->` or `<!-- addressed:c
    1. Read the comment and understand the requested change
    2. Make the fix as a fixup commit: `~/.cursor/commands/lint-commit.sh --fixup <hash> [files...]`
    3. Push the fixup to the branch
-   4. **Remove this PR from the merge set** — it needs re-review after the fixup
-   5. Continue with remaining PRs that have no outstanding comments
-   6. Report addressed PRs to the user at the end of the workflow
+   4. Reply on the PR thread explaining what was fixed (1 sentence, factual). Use `gh pr comment <number> --repo EdgeApp/<repo> --body "..."` for top-level comments, or reply to the specific thread if the feedback was inline.
+   5. **Remove this PR from the merge set** — it needs re-review after the fixup
+   6. Continue with remaining PRs that have no outstanding comments
+   7. Report addressed PRs to the user at the end of the workflow
 
 **Do NOT block the rest of the flow** for PRs with comments.
 </sub-step>

.cursor/rules/answer-questions-first.mdc

@@ -13,5 +13,6 @@ Before using any code editing tools, scan the user's message for `?` characters.
 If questions are detected:
 
 1. Read `~/.cursor/commands/q.md` and follow its workflow to answer every question.
-2. Do **not** edit files, create files, or run mutating commands until the user responds.
-3. Only proceed with implementation after the user permits it in a follow-up message.
+2. **Workflow context**: If a command or skill was invoked earlier in this conversation, note which one. When a question or critique references agent behavior from that execution, load the command/skill definition before answering and evaluate whether the command should have governed that behavior. If it should have but didn't, that's a workflow gap — treat it as the primary concern per `fix-workflow-first.mdc`.
+3. Do **not** edit files, create files, or run mutating commands until the user responds.
+4. Only proceed with implementation after the user permits it in a follow-up message.

.cursor/rules/typescript-standards.mdc

@@ -209,4 +209,20 @@ const getStyles = cacheStyles((theme: Theme) => ({
 Note: This is an architectural change. If the file has many `styled()` usages, flag to user rather than refactoring inline.
 </pattern>
 
+<pattern id="cleaner-error-matching" rule="@typescript-eslint/use-unknown-in-catch-callback-variable">When catching unknown errors that need property inspection, use `cleaners` instead of type assertions.
+❌ `const err = e as { code?: string; message?: string }`
+✅ Define a cleaner and use `asMaybe`:
+```ts
+const asFooError = asObject({
+  code: asValue(FOO_CODE),
+  message: asOptional(asString, '')
+})
+const fooError = asMaybe(asFooError)(e)
+if (fooError != null) { ... }
+```
+For generic error message extraction:
+❌ `err.message ?? ''` (unsafe on `unknown`)
+✅ `e instanceof Error ? e.message : String(e)`
+</pattern>
+
 </lint-fix-patterns>

.cursor/skills/author/SKILL.md

@@ -92,13 +92,21 @@ When revising an existing command, **every item below is mandatory** — not a s
 
 1. Read the full file before making changes
 2. Check for duplicated logic across other commands — consolidate if found
-3. Verify step ordering matches the agent's decision flow
-4. Ensure examples are brief and generic (no real repo names, PR numbers, or user data)
-5. Check that escape hatches exist for ambiguous cases
-6. Confirm companion scripts match the `.md` expectations
-7. Convert markdown-structured commands to XML format (this is the most commonly skipped item — `##` headers and bullet lists must become `<goal>`, `<rules>`, `<step>` tags)
-8. Apply all current authoring principles (rules-first, scripts-over-reasoning, batch-tool-calls, etc.) even if the original command predates them
-9. If the command may run on smaller/faster models, apply `<small-model-conventions>` — especially `file-over-args`, `inline-guardrails`, and `verbatim-bash`
+3. **Check behavioral dependencies**: Search for other commands, skills, and rules that perform similar operations or share domain overlap with the one being edited. If command A has a step that is a lightweight version of command B's core behavior (e.g., `/pr-land` addressing comments vs `/pr-address`), verify that A's step is consistent with B's rules — missing rules in A are likely bugs.
+   - Extract domain-specific verbs and nouns from the step being edited (e.g., a step about handling PR comments yields: `comment`, `reply`, `resolve`, `address`, `fixup`, `thread`)
+   - Search each term across commands, skills, and rules:
+   ```bash
+   rg -l "<term>" ~/.cursor/commands/*.md ~/.cursor/skills/*/SKILL.md ~/.cursor/rules/*.mdc
+   ```
+   - Read any hits that share domain overlap and check for consistency
+   - If overlap is found, evaluate whether to consolidate per the `dry` principle: can A reference B's rules or a shared file instead of reimplementing? Propose consolidation to the user when the shared logic is non-trivial.
+4. Verify step ordering matches the agent's decision flow
+5. Ensure examples are brief and generic (no real repo names, PR numbers, or user data)
+6. Check that escape hatches exist for ambiguous cases
+7. Confirm companion scripts match the `.md` expectations
+8. Convert markdown-structured commands to XML format (this is the most commonly skipped item — `##` headers and bullet lists must become `<goal>`, `<rules>`, `<step>` tags)
+9. Apply all current authoring principles (rules-first, scripts-over-reasoning, batch-tool-calls, etc.) even if the original command predates them
+10. If the command may run on smaller/faster models, apply `<small-model-conventions>` — especially `file-over-args`, `inline-guardrails`, and `verbatim-bash`
 </revision-checklist>
 
 <companion-scripts>