fix(auth): reconcile Codex auth sync and runtime config with main

.agents/skills/ai-slop-detector/SKILL.md

@@ -2,10 +2,12 @@
 name: ai-slop-detector
 description: Use when the user asks to audit, score, humanize, or fix prose for AI-writing signals or readability. Produces separate AI-slop and comprehension findings with actionable edits.
 tags: [writing, editing, readability, ai-detection]
+capability: Audit prose for AI-writing signals and reader comprehension, then return prioritized edits.
 triggers:
   - humanize this
   - AI slop check
   - does this sound AI-written
+  - read like it was generated by AI
   - readability audit
 ---
 

.agents/skills/hydra/SKILL.md

@@ -2,6 +2,7 @@
 name: hydra
 description: Use when the user asks for Hydra, a deep multi-perspective code review, architecture or security tradeoffs, blind spots, or iterative re-review. Avoid for simple fixes and lookups.
 tags: [code-review, architecture, security, multi-perspective]
+capability: Run a multi-perspective review council for architecture, security, and hidden risks.
 triggers:
   - run Hydra
   - Hydra review

.agents/skills/reddit-skills/SKILL.md

@@ -2,10 +2,15 @@
 name: reddit-skills
 description: Use when the user asks to operate Reddit across authentication, search, publishing, comments, votes, saves, subreddit analysis, or multi-step content workflows.
 tags: [reddit, automation, social-media]
+capability: Coordinate Reddit search, publish, comment, login, vote, and save tasks across multiple skills.
 triggers:
   - use Reddit
   - operate Reddit
   - Reddit workflow
+  - several Reddit actions
+  - Search Reddit, then comment
+  - Browse a subreddit and save a post
+  - Check Reddit login and publish a link post
 version: 1.0.0
 metadata:
   openclaw:

.agents/skills/reddit-skills/skills/reddit-auth/SKILL.md

@@ -2,6 +2,7 @@
 name: reddit-auth
 description: Use when the user asks to check Reddit login status, authenticate a Reddit session, or log out. Handles authentication state only.
 tags: [reddit, authentication]
+capability: Inspect or change Reddit authentication state without performing content operations.
 triggers:
   - am I logged in to Reddit
   - check my Reddit login

.agents/skills/reddit-skills/skills/reddit-content-ops/SKILL.md

@@ -2,6 +2,7 @@
 name: reddit-content-ops
 description: Use when the user asks for multi-step Reddit operations such as subreddit analysis, trend tracking, content strategy, publishing, or engagement campaigns.
 tags: [reddit, content-strategy, analysis, automation]
+capability: Execute compound Reddit research, strategy, publishing, and engagement workflows.
 triggers:
   - analyze subreddit trends
   - Reddit content strategy

.agents/skills/reddit-skills/skills/reddit-explore/SKILL.md

@@ -2,6 +2,7 @@
 name: reddit-explore
 description: Use when the user asks to search Reddit, browse subreddits, inspect posts, track discussions, or view Reddit user profiles. Read-only discovery and research.
 tags: [reddit, search, research]
+capability: Perform read-only Reddit discovery across posts, subreddits, discussions, and profiles.
 triggers:
   - search Reddit
   - browse subreddits

.agents/skills/reddit-skills/skills/reddit-interact/SKILL.md

@@ -1,10 +1,12 @@
 ---
 name: reddit-interact
-description: Use when the user asks to comment, reply, upvote, downvote, or save Reddit content. Handles social interactions but does not publish new posts.
+description: Use when the user asks to comment, reply, upvote, downvote, or save Reddit content. Handles social interactions. NOT for publishing new posts.
 tags: [reddit, engagement, social-media]
+capability: Perform Reddit comments, replies, votes, and saves without creating new posts.
 triggers:
   - post a comment on Reddit
   - upvote that Reddit post
+  - downvote this Reddit post
   - save Reddit content
 version: 1.0.0
 metadata:

.agents/skills/reddit-skills/skills/reddit-publish/SKILL.md

@@ -1,7 +1,8 @@
 ---
 name: reddit-publish
-description: Use when the user asks to publish, submit, or share a text, link, or image post on Reddit. Handles new posts rather than comments or votes.
+description: Use when the user asks to publish, submit, or share a text, link, or image post on Reddit. NOT for blogs, company websites, comments, or votes.
 tags: [reddit, publishing, social-media]
+capability: Create new Reddit text, link, or image posts without handling comments or votes.
 triggers:
   - create a new text post on Reddit
   - publish a Reddit post

.claude/reviews/pr-128-review.md

@@ -0,0 +1,81 @@
+# PR Review: #128 — lint: stop --fix from destroying inline code and allowed-tools lists
+
+**Reviewed**: 2026-08-07
+**Author**: NagyVikt
+**Branch**: fix/skill-linter-fixer-data-loss → main
+**Decision**: APPROVE (after two defects found here were fixed in f73481d4)
+
+> **Reviewer independence caveat.** This is a self-review: the same session
+> authored the PR. Shared blind spots are not excluded. Both findings below
+> were confirmed by running code, not by reading it, which is the part of the
+> review that does not depend on the reviewer's priors.
+
+## Summary
+
+Repairs two content-destroying defects in `cue lint-skill --fix`, both
+confirmed against the 453-skill corpus. The first review pass of this PR found
+two further defects in its own change; both are fixed and covered by tests.
+
+## Findings
+
+### CRITICAL
+None.
+
+### HIGH
+
+**H1 — space-separated bare names regressed** (`src/lib/skill-linter.ts`, `readAllowedTools`) — FIXED in f73481d4
+
+The original split on `/[,\s]+/`. The new `readAllowedTools` split on commas
+only, so a space-separated value collapsed into one bogus entry:
+
+```
+allowed-tools: nmap curl   ->   Bash(nmap curl:*)      # regressed
+                           ->   Bash(nmap:*), Bash(curl:*)   # expected
+```
+
+Verified by executing `applyFixes` on both shapes. Fixed with `splitToolNames`,
+which restores the old behaviour but keeps a part containing `(` whole so
+`Bash(git diff:*)` is not torn in half — a case the original only survived by
+early-returning on any wrapped value, which also meant it fixed nothing in a
+mixed list. Corpus impact: none — zero skills carry a bare multi-word value.
+
+### MEDIUM
+
+**M1 — `fixedNames` computed and never read** (`src/lib/skill-linter.ts`) — FIXED in f73481d4
+
+The `fix` closure recomputes the same map from a fresh parse, so the outer
+binding was dead. `bun run lint` flags it ("This variable fixedNames is
+unused"); CI passed only because unused variables are a warning, not an error.
+Removed. The wrap predicate had been written out three times, which is how the
+dead copy went unnoticed; it is now `isWellFormedTool` + `wrapTool`.
+
+### LOW
+
+**L1 — `readAllowedTools` stops at the first non-item line.** A comment
+interleaved in a block sequence would truncate the parse. No occurrence in the
+corpus; noting it rather than speculatively handling it.
+
+## Validation Results
+
+| Check | Result |
+|---|---|
+| Type check (`bun run typecheck`) | Pass |
+| Lint (`bun run lint`) | Pass, and the warning this PR had added is gone |
+| Tests (`bun test`) | Pass — 3091 pass / 1 skip / 0 fail |
+| Unit (`skill-linter.test.ts`) | Pass — 94 tests |
+| CI on GitHub | test, lint, Profiles e2e ×2, configured — all pass |
+
+## Behavioural verification (beyond the test suite)
+
+| Property | Result |
+|---|---|
+| inline code spans lost over 453 skills | 0 (was 723 across 178 files) |
+| `allowed-tools` list → scalar degradations | 0 |
+| frontmatter unparseable / `name:` / `description:` / fences altered | 0 / 0 / 0 / 0 |
+| new `Bash(Bash:*)` introduced | 0 (47 pre-existing, unchanged) |
+| reproduces opencue/skills#21 on its own base | byte-identical |
+
+## Files Reviewed
+
+- `src/lib/skill-linter.ts` — Modified
+- `src/lib/skill-linter.test.ts` — Modified