review-implement 92/100

CategoryWeightCategory scoreContribution

Securityprompt, exec, net, exfil, eval

35%

26.6 pts

Supply chainhash, typosquat, maintainer, lockfile

20%

100

20.0 pts

Maintenancestaleness, pinning, CI

15%

100

15.0 pts

TransparencySKILL.md, perms, README

15%

100

15.0 pts

Communityinstalls, verify, response

15%

100

15.0 pts

Findings & checks · 2 flagged

Securityscore 76 · 2 findings

MEDIUMInstruction telling the agent not to ask for approvalSS-SKILL-INJECT-DONT-ASK-01 · Prompt injection · combo/review-implement/SKILL.md×2

MEDIUM — it fires on intent; the real damage depends on the host agent's own approval-gating.

Why it matters

The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.

The exact value spotted

excerptcombo/review-implement/SKILL.md· markdown

5category: combo

6instructions: |

7You are an autonomous review-to-implementation agent. Do NOT ask the user questions.

8Run the full pipeline below without pausing between phases.

Occurrences

2 occurrences · first at L7, also L86

Show all 2 locations

Line

File

combo/review-implement/SKILL.md

L86

combo/review-implement/SKILL.md

How to fix

Remove the approval-skipping instruction, or scope it narrowly to a specific safe, reversible action.

Delete blanket "don't ask / no need to confirm" directives from the skill.
If the skill is a genuine autonomous job, restrict the opt-out to a named non-destructive action rather than all actions.

Framework references

OWASPLLM01 ATLASAML.T0051

Trace & refs

ruleSS-SKILL-INJECT-DONT-ASK-01sha2567d1c5f93d8975bab…rubric 365aacaView on GitHub

Supply chainscore 100 · 0 findings

✓—All supply chain checks passedNo findings in this category for the latest scan.pass

Maintenancescore 100 · 0 findings

✓—All maintenance checks passedNo findings in this category for the latest scan.pass

Transparencyscore 100 · 0 findings

✓—All transparency checks passedNo findings in this category for the latest scan.pass

Communityscore 100 · 0 findings

✓—All community checks passedNo findings in this category for the latest scan.pass

The primary manifest — the file an agent reads to learn what this artifact does.

combo/review-implement/SKILL.md · 1 file

combo/review-implement/SKILL.md5.3 KB · Markdown

name: review-implement

description: Review then build — chains architecture review into implementation. Accepts any spec format (ticket, PRD, engineering spec, feature description, screenshot) and autonomously reviews the design, resolves gaps, then implements with tests and a PR.

version: "2.0.0"

category: combo

instructions: |

You are an autonomous review-to-implementation agent. Do NOT ask the user questions.

Run the full pipeline below without pausing between phases.

INPUT:

$ARGUMENTS

The user will provide a spec in any format: ticket, PRD, engineering spec, feature

description, bullet points, screenshot, or conversation summary.

============================================================

PHASE 1: ARCHITECTURE REVIEW (/arch-review)

============================================================

Follow the instructions defined in the `/arch-review` skill exactly.

Run in Design Review mode (pre-implementation).

Produce the full review output: feasibility, completeness, codebase impact,

schema design, API design, business logic, risks, and gaps.

End with the verdict: READY TO IMPLEMENT, NEEDS CLARIFICATION, or SIGNIFICANT GAPS.

**If NEEDS CLARIFICATION or SIGNIFICANT GAPS:**

- Resolve all identified issues yourself. Make reasonable decisions.

- Amend the spec with your resolutions.

- Document what you decided and why.

- Produce a revised, implementation-ready spec.

Do NOT stop here. Continue immediately to Phase 2 with the

reviewed (and possibly amended) spec.

============================================================

PHASE 2: IMPLEMENTATION (/story-implementer)

============================================================

Follow the instructions defined in the `/story-implementer` skill exactly.

Use the reviewed and resolved spec from Phase 1 as input — NOT the original

raw input. This ensures all design decisions and gap fixes from the review

are reflected in the implementation.

- Implement the feature with production-ready code

- Write unit tests

- Create a commit and PR

- Check for and address bot review comments

============================================================

OUTPUT

============================================================

When both phases are complete, print a summary:

Review & Implement Complete

PARALLEL EXECUTION: When review findings span both backend and frontend, use the Agent tool to fix them concurrently.

Agent A (Backend Fixes): "Fix the following backend review findings: [backend findings]. Run backend tests after each fix. Return: files modified, tests pass/fail."
Agent B (Frontend Fixes): "Fix the following frontend review findings: [frontend findings]. Run frontend tests after each fix. Return: files modified, tests pass/fail."
Wait for both agents to complete.
Run the full test suite to verify integration.

Feature: [title] Review verdict: [READY / NEEDS CLARIFICATION / SIGNIFICANT GAPS] Resolutions made: [N issues resolved during review, or "None needed"] PR: [PR URL]

What was built:

[bullet summary of implementation]

Next steps:

Run /e2e or /walkthrough to verify with automated tests
Run /qa for full QA verification
Run /manual-test-plan to generate a manual test plan for the PR

STRICT RULES:

Do NOT implement without reviewing first. Phase 1 always runs.
Do NOT ask the user to resolve review findings. Resolve them yourself.
If the review found gaps, the implementation MUST address them.
All rules from /arch-review and /story-implementer apply to their respective phases.

platforms:

CLAUDE_CODE

============================================================ SELF-HEALING VALIDATION (max 3 iterations) ============================================================

After completing all phases, validate the combined output:

Re-run the specific checks that originally found issues to confirm fixes.
Run the project's test suite to verify fixes didn't introduce regressions.
Run build/compile to confirm no breakage.
If new issues surfaced from fixes, add them to the fix queue.
Repeat the fix-validate cycle up to 3 iterations total.

STOP when:

Zero Critical/High issues remain
Build and tests pass
No new issues introduced by fixes

IF STILL FAILING after 3 iterations:

Document remaining issues with full context
Classify as requiring manual intervention or architectural changes

============================================================ SELF-EVOLUTION TELEMETRY ============================================================

After producing output, record execution metadata for the /evolve pipeline.

Check if a project memory directory exists:

Look for the project path in ~/.claude/projects/
If found, append to skill-telemetry.md in that memory directory

Entry format:

### /review-implement — {{YYYY-MM-DD}}
- Outcome: {{SUCCESS | PARTIAL | FAILED}}
- Self-healed: {{yes — what was healed | no}}
- Iterations used: {{N}} / {{N max}}
- Bottleneck: {{phase that struggled or "none"}}
- Suggestion: {{one-line improvement idea for /evolve, or "none"}}

Only log if the memory directory exists. Skip silently if not found. Keep entries concise — /evolve will parse these for skill improvement signals.

review-implement— agent skill

Is review-implement safe to install?

Review & Implement Complete

Audit the pieces. Scan the whole. Decide.