AgentStack
Browse Sign in
Browse Why AgentStack Sell Docs
Sign in
SKILL verified Apache-2.0 Self-run

Safe Remediation

skill-alnah-agent-safe-remediation · by alnah

Use this skill when applying recommendations from an audit, refactor plan, migration plan, incident review, security review, or architecture review while minimizing regressions and unintended side effects.

No reviews yet
0 installs
31 views
0.0% view→install

Install

$ agentstack add skill-alnah-agent-safe-remediation

✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access No
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets No
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

View the full security report →

Verified badge

Passed review? Show it. Paste this badge into your README, it links to the public security report.

AgentStack Verified badge Links to your public security report.
[![AgentStack Verified](https://agentstack.voostack.com/badges/verified.svg)](https://agentstack.voostack.com/security/report/skill-alnah-agent-safe-remediation)

Reliability & compatibility

Security review passed
0 installs to date
no reviews yet
2mo ago

Declared compatibility

Claude CodeClaude Desktop

Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.

Preview Execution monitoring

We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.

How agent discovery & health will work →
Are you the author of Safe Remediation? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

Safe Remediation

Goal: apply a finding with minimum iatrogenic risk. Treat every remediation as a risky intervention until validated.

Core rules

  • Do not apply audit recommendations blindly.
  • Prefer smallest useful change over broad cleanup.
  • Separate behavior change from refactor.
  • Preserve public contracts unless explicit approval exists.
  • Add or identify safety tests before modifying risky code.
  • Keep rollback possible until production confidence exists.
  • Stop if blast radius, validation, or rollback path is unknown.

Workflow

  1. Clarify the finding
  • Identify the exact risk, affected files, user impact, and maintainer impact.
  • Reject purely aesthetic changes unless they unblock a concrete risk.
  • Define the intended outcome in one sentence.
  1. Classify the change
  • Refactor only: observable behavior must not change.
  • Bug fix: behavior changes only for specified failure cases.
  • API or contract change: compatibility and consumers are primary risk.
  • Data or schema migration: rollback and forward compatibility are primary risk.
  • Config or deploy change: runtime blast radius and observability are primary risk.
  • Dependency upgrade: transitive behavior and security are primary risk.
  1. Establish a safety baseline
  • Run current validation commands first when safe.
  • Add characterization tests for current behavior if coverage is weak.
  • Add contract, golden, integration, or regression tests around the touched boundary.
  • Record known failures separately from new failures.
  1. Analyze blast radius
  • Inspect imports, call sites, public API users, schemas, config keys, jobs, queues, and external resources.
  • Identify high fan-in modules and hidden shared state.
  • Identify concurrency, lifecycle, and cleanup paths.
  • Decide whether the change needs a feature flag, adapter, compatibility shim, or staged migration.
  1. Plan small reversible steps
  • PR 1: tests or observability only.
  • PR 2: compatibility scaffolding or seam.
  • PR 3: implementation change behind the seam or flag.
  • PR 4: rollout or default switch.
  • PR 5: cleanup after confidence.
  • Keep each step independently reviewable and revertible.
  1. Implement with guardrails
  • Make one conceptual change at a time.
  • Run targeted tests after each meaningful edit.
  • Avoid opportunistic formatting, renames, dependency upgrades, or unrelated cleanup.
  • Preserve old behavior behind compatibility code when external consumers may exist.
  1. Validate
  • Run targeted tests.
  • Run broader tests relevant to the blast radius.
  • Run lint, typecheck, build, race/concurrency checks, or migration dry-runs when relevant.
  • For runtime changes, require metrics by version or cohort: errors, latency, saturation, correctness, and business-critical signals.
  1. Rollout and rollback
  • Define rollback before merge.
  • Prefer canary, dark launch, or feature flag for risky runtime changes.
  • For database changes, use expand-contract: add compatible schema, deploy compatible code, switch use, then remove old schema later.
  • Test rollback periodically if the system supports it.

Stop conditions

Stop and ask for direction if any condition is true:

  • No safe validation command exists for the touched behavior.
  • Behavior baseline is unknown and cannot be characterized cheaply.
  • Rollback would be impossible or data-destructive.
  • Public API, schema, or config compatibility would break unexpectedly.
  • The change combines refactor, behavior change, and dependency upgrade.
  • The proposed edit touches many unrelated modules.
  • The audit recommendation lacks evidence or success metric.

Output format

Use this concise format unless the user asks otherwise:

Finding: 
Change type: 
Blast radius: 
Safety baseline: 
Plan:
1. 
2. 
Validation:
- 
Rollback:
- 
Stop conditions: 

Checklist before final answer

  • State files changed.
  • State commands run and result.
  • State commands skipped and why.
  • State remaining risks.
  • Do not claim safety beyond evidence.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

  • Author: alnah
  • Source: alnah/agent
  • License: Apache-2.0
  • Homepage: https://alnah.me

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet, be the first.

Versions

  • v0.1.0 Imported from the upstream source.