agentsclimarketplace

Existing project audit

Skill ahtishamshahzad/agent_dev_flow/.ai/skills/existing-project-audit

Use when a codebase already exists, before proposing any changes. It inventories structure, stack, applications present, tests, security posture, and implementation state so planning is grounded in reality rather than assumptions.From its SKILL.md

Install
npx -y skills add ahtishamshahzad/agent_dev_flow --skill existing-project-audit

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 25 days oldThe repository was created 25 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

4.0 KB, 816 tokens by cl100k_base, as published. Nobody here has run it

Existing Project Audit

Purpose

Understand what already exists before changing it. Produce an accurate picture of the repository — structure, stack, applications, tests, security posture, and state — that requirements, application, and stack decisions can rely on.

When to Use

  • Any request against an existing codebase (enhancement, feature, bug, refactor, migration, review).
  • Before recommending changes or a stack for an existing repo.
  • Not for greenfield work (record "greenfield" and skip).

Inputs

  • Read access to the repository.
  • The classified request (to scope the audit).

Discovery Questions

  • Which parts of the repo are in scope for this request?
  • Are there known problem areas, recent incidents, or fragile modules?
  • What is the current build/test/deploy story?

Responsibilities

  • Inventory structure (apps, packages, modules, entry points).
  • Detect the stack actually in use (frameworks, DB, data layer, testing tools).
  • Identify applications present (mobile/web/dashboard/backend/worker/etc.).
  • Assess tests (levels, tools, coverage signal).
  • Assess security posture at a high level (secrets, auth, exposure) — flag, don't fix here.
  • Determine implementation state (what's done, in-progress, stubbed).
  • Record findings; do not edit code during the audit.

Required Workflow

  1. Scope the audit to the request.
  2. Map structure and entry points.
  3. Detect stack and applications from manifests/config/code.
  4. Survey tests and CI.
  5. Note security-relevant signals (hand deep analysis to security-review).
  6. Summarize state and risks in ../../projects/current/ / ../../generated/audits/.

Decision Rules

  • Report what the repo proves, separate from what you infer.
  • If the request is small, scope the audit to the affected area — don't audit the whole repo needlessly.
  • Deep security/performance analysis is delegated (security-review, performance-review), not done here.

Rules

  • Read-only: no edits during the audit.
  • Do not print discovered secret values — flag location/type, redacted (../../system/SECURITY_RULES.md).
  • Ground later stages in findings, not guesses.

Anti-Patterns

  • Proposing changes before understanding the current state.
  • Auditing the entire monorepo for a one-file fix.
  • Treating inferences as confirmed facts.

Validation Checklist

  • Structure and entry points mapped.
  • Stack and applications detected.
  • Tests and CI surveyed.
  • Security signals flagged (redacted).
  • Implementation state and risks recorded.
  • No code edited.

Definition of Done

A recorded audit (scoped to the request) covering structure, stack, applications, tests, security signals, and state — with confirmed vs inferred separated — ready to inform requirements, applications, and stack.

Related Skills

request-classification, requirements-analysis, application-selection, stack-recommendation, dependency-audit, environment-audit, security-review, performance-review.

Related Knowledge

../../knowledge/ (existing architecture notes, if any).

Related References

None typically; a stack reference folder only if an unfamiliar framework is present.

Context Loading Guidance

  • Requires: repository read access, the classified request.
  • Does not require: the full stack decision space, unrelated references, or every skill.
  • May load: dependency-audit, environment-audit for deeper inventory; security-review for posture.
  • Stop when: the scoped audit is recorded.

Token Efficiency Guidance

Read entry points, manifests, and config first; sample representative files rather than reading everything. Summarize into findings; don't paste large files into the record.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 326,782. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.