Harden My Repo Report

97/100

Evaluation coverage: 88% · Profile: standard

Bounded claim: This is a local configuration-readiness review, not a security audit, certification, or guarantee.

Generated 2026-07-16T18:22:35.445Z · Tool 1.0.0 · Checks 2026-07-16.1

Scope

10 eligible text files read. 0 symlinks and 0 binaries skipped. No network requests or project commands were run.

Score uses applicable pass/warn/fail checks only (pass=100%, warn=50%, fail=0%) weighted by impact. Manual and not-applicable checks do not increase the score; coverage shows how much applicable weight was evaluated.

AR-001 — Repository-level agent instructions

PASS · high severity · small effort

Impact

Without repository-specific instructions, an agent must infer boundaries, commands, and conventions.

Remedy

Add a concise root instruction file covering scope, forbidden actions, verification commands, and escalation points.

Verification

Confirm a root CLAUDE.md, AGENTS.md, or Copilot instruction file exists and is reviewed by the team.

Evidence

Unknowns

AR-002 — Machine-readable agent settings

PASS · medium severity · small effort

Impact

Absent settings make behavior depend on user or machine defaults.

Remedy

Add reviewed shared project settings appropriate to the selected operating profile.

Verification

Parse the settings as JSON and inspect effective permissions in the agent runtime.

Evidence

AR-003 — Permissions align with standard profile

PASS · high severity · medium effort

Impact

Permissions that are broader than intended can allow destructive or externally visible actions without the expected checkpoint.

Remedy

Define explicit deny rules for destructive commands and sensitive paths.

Verification

Review the effective permission policy and exercise representative allowed, denied, and confirmation-required commands.

Evidence

Unknowns

AR-004 — Destructive-command guardrail

PASS · high severity · medium effort

Impact

Agents can propose or run commands that delete work, rewrite history, publish packages, or mutate data.

Remedy

Add a tested pre-tool hook or equivalent policy that blocks destructive command families and fails closed.

Verification

Run safe positive and bypass-oriented negative tests against the guardrail.

Evidence

Unknowns

AR-005 — Secret-leak prevention control

PASS · high severity · medium effort

Impact

Credentials can be exposed through commits, prompts, logs, or generated patches.

Remedy

Add a maintained pre-commit or CI secret detector and document credential handling.

Verification

Use a fake canary value to confirm the control blocks or flags it without printing the value.

Evidence

Unknowns

AR-006 — Documented verification commands

MANUAL · medium severity · small effort

Impact

Without known verification commands, changes may be accepted without meaningful feedback.

Remedy

Document deterministic test, lint, typecheck, and build commands in the agent instructions.

Verification

Run the documented commands in the project’s supported environment; this tool intentionally does not execute them.

Unknowns

AR-007 — CI agent operating boundary

NA · high severity · medium effort

Impact

Unbounded CI automation can act with repository tokens or mutate pull requests without an explicit operating contract.

Remedy

Define least-privilege CI permissions, trigger boundaries, concurrency controls, and artifact/log handling.

Verification

Review CI token permissions and test the workflow on an unprivileged fixture repository.

Unknowns

AR-008 — Context and ownership boundaries

WARN · medium severity · small effort

Impact

Large or ambiguous repositories increase the chance that an agent edits generated, vendored, or team-owned areas.

Remedy

Document generated paths, ownership boundaries, sensitive areas, and where nested instructions apply.

Verification

Ask maintainers to review the documented boundaries against current repository layout.

Unknowns

AR-009 — Optional secret-shaped value scan

NA · high severity · small effort

Impact

Secret-shaped values in eligible scanned text can indicate accidental credential storage.

Remedy

Run with --secret-scan when authorized; treat results as indicators, not proof.

Verification

Confirm findings without copying values. If real, rotate through the issuing provider and inspect history separately.

Unknowns

AR-010 — Repository scan coverage limits

PASS · medium severity · small effort

Impact

Files outside scan limits can contain relevant configuration, so conclusions may be incomplete.

Remedy

Review excluded and skipped paths manually; narrow the repository or use explicit --exclude patterns to make coverage intentional.

Verification

Inspect the coverage section and account for each skipped category.