scriptedalchemy/tracedecay/reviewing-changes
Claude Code and agent skill by scriptedalchemy/tracedecay
- community
Checking that the SKILL.md file is still in the repository.
Use when the user asks to review, summarize, or assess the risk of a PR, branch, or working-tree diff. Load this skill before the first tool call.
Install
npx vskill@latest install scriptedalchemy/tracedecay/reviewing-changesSKILL.md
Reviewing changes
Choose the evidence that answers the review question. Start with an orientation packet, read the relevant code, and deepen only where behavior or risk remains uncertain.
Choose the comparison
- For a PR or committed ref range, call
tracedecay_pr_contextfor its compact
briefing: description, retained discussion and checks, change map, and ranked reading entrypoints. Hosted evidence is retained by default. Set refresh_github: true only when a provider refresh is useful; it refreshes the configured matching PR within the request deadline. CI remains retained.
- For an uncommitted change summary, call
tracedecay_commit_context.
staged_only: true compares pinned HEAD with immutable index blobs. The default compares HEAD with captured working-tree bytes, including untracked files. This is the net result: a staged edit undone in the working tree does not appear in the default summary. Read its comparison and source coverage.
- For structured file/hunk metadata, use
tracedecay_git_diff. Its
working_tree scope compares index to worktree; staged compares HEAD to index. Read raw patches with shell git diff or git diff --cached.
- Use
tracedecay_diff_contextfor prospective context on specified files. It
includes unchanged indexed declarations, possible consumers, and test candidates. Use PR or commit context when exact changed declarations matter.
Read the evidence
Read the briefing's entrypoints with their recorded revision and source span. For deleted or non-head source, use tracedecay_source_lines with the full commit revision, repository-relative file, and one-based inclusive start_line/end_line. Continue at returned next_line with the same revision, file, and original end line; the reader returns whole lines. Historical reads use the normal admitted source-service lifecycle and configured path policy. If a read is unavailable, report the limitation; an authorized shell git show '<commit-oid>:path/to/file' can inspect the same committed source. Staged blob IDs are provenance, not historical commit selectors; use git diff --cached for their patch text.
When an exact-head graph match supplies a node_id, pass it directly to source_body, impact, or test_map. Graph links cannot prove unindexed or external consumers safe. Use view: "symbols" only when exact symbol pagination helps the task. The brief's token budget is approximate; inspect omission counts and over_budget before assuming its sections are exhaustive.
Expand relevant discussion through supplied provider URLs. A retained body_anchor identifies provenance; it is not a response-cache retrieve handle. feedback_expand requires a server-minted finding expansion handle. Retained checks are observations, not a live inventory of required checks.
Deepen on risky changes through callers, impact, and test links. Confirm each candidate finding against concrete behavior and a reachable caller. A scan hit is a lead: test unwraps and unsafe blocks are not defects by themselves. For dead-code or ship-risk work, see safety audit.
A review alone grants no write authority. Honor writes already authorized by the user; otherwise read write authority before Git apply, native integration, or worktree cleanup. Structural test selection belongs to assessing-impact; structural mutation belongs to editing-safely.
Report: actionable findings with location, failure mode, and evidence, plus material coverage gaps. Distinguish executed tests from structural test links. Do not apply changes without write authority.
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewNot runA model review is scheduled for the next rescan.
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingplugin/skills/reviewing-changes/SKILL.md
Popularity
- Repository stars
- 83
- Repository forks
- 7
- 7-day trend score
- 27
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/scriptedalchemy/tracedecay/reviewing-changes)