Give a code-review or PR-description flow a structural head start before reading a diff line by line. Trigger when the user asks for a PR description, a pre-merge summary, or invokes a code-review flow (this repo's `open-agent-hub:review`/`/review`, or an equivalent in another environment) against a diff in a repo with a `.dekko/` directory.
Scanned 9/3/2026
Install to Claude Code
npx -y skills add aahlijia/dekko --skill dekko-review-context --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Dekko Review Context?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/aahlijia-dekko-review-context)More formats (shields.io, HTML) on the badges page.
---
name: dekko-review-context
description: Give a code-review or PR-description flow a structural head start before reading a diff line by line. Trigger when the user asks for a PR description, a pre-merge summary, or invokes a code-review flow (this repo's `open-agent-hub:review`/`/review`, or an equivalent in another environment) against a diff in a repo with a `.dekko/` directory.
---
# Structural context before reviewing a diff
A reviewing agent that reads a diff cold misses what a call graph
already knows: what else calls the changed code, what tests should
run, and where dekko's own resolver is unsure. `workset`,
`impacted_tests`, and `check_ambiguous` already answer these — this
skill composes them into review context, run *before* the diff itself,
so the read is top-down (structure first, then lines) instead of
bottom-up.
## Procedure
| Need | Use |
|---|---|
| What changed and what calls it | `mcp__dekko__workset [rev]` (or `dekko workset [REV]`) |
| What tests should run | `mcp__dekko__impacted_tests [rev]` (or `dekko affected [REV]`) |
| Where the resolver itself is unsure | `mcp__dekko__check_ambiguous` (or `dekko` equivalent) |
1. **Determine the diff scope.** A git rev range, or the same default
`workset`/`impacted_tests` already use: the commit the map was
generated at, else `HEAD`. Don't ask the user to specify this
unless the default is ambiguous for the request.
2. **Call `workset [rev]`.** Bundles touched-file outlines plus
call-graph packs for the most central touched symbols under one
token budget — this becomes the "what changed and what calls it"
section of the review context.
3. **Call `impacted_tests [rev]`.** Reverse call-graph reachability
from the changed symbols, more reliable than grepping test files
for the changed symbol's name — this becomes the "what tests
should run" section.
4. **Call `check_ambiguous`.** It's repo-wide, not diff-scoped, so
cross-reference its top-colliding names/files against the touched
symbols from step 2 yourself. Any overlap gets flagged explicitly
in the review output: "dekko's resolver disclaims confidence near
`<file>`/`<name>` — verify this part of the diff by hand," rather
than silently trusting a clean-looking diff in a resolver-weak
area.
5. **Fold all three into the review/PR-description prompt before
reading the diff line by line.** Present structural context first,
then read the actual diff — not appended as an afterthought.
## Forward-looking note (issue #14)
Steps 2-4 above are a manual composition of tools that exist today,
built because a first-class `dekko review [REV1] [REV2]` command
(issue #14) doesn't exist yet. When #14 lands, collapse steps 2-4 into
a single call to that command instead — this skill's procedure should
shrink at the same time #14 ships, not stand as a permanent example of
the pre-#14 workaround shape.
## Boundaries
- This does not replace `open-agent-hub:review` or an equivalent
review flow — it's a context *supplement* that fires before or
alongside one, the same way `dekko-orient` supplements normal
navigation without owning it. It composes into whatever review/PR
flow is already invoked, not a review output format of its own.
- Structural context, not a substitute for reading the diff — use it
to prime the read, not to skip it.
- `check_ambiguous` is repo-wide; the cross-reference against touched
symbols is reasoning you do, not a tool that already joins the two.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!