Index of every page on this site: llms.txt
All skills
Run it yourself

Read-only audit of an existing Bolt CMS 6 project against every bolt-skills skill. Profiles the project, runs all 20 verification scripts in workflow order, compares the project's files with the skills' examples, then reviews each skill one by one (logs, SKILL.md rules, references) and writes a prioritized AUDIT.md report — defects vs. convention deviations, severity, evidence and the skill step that fixes each finding. Use when asked to audit, review, check or assess an existing Bolt site, or before taking over or upgrading one.

  • 5 files
  • 62.3 KB
  • Updated 2 weeks ago
  • GitHub
Use this Skill: https://skilld.dev/gh/vondry/bolt-skills/bolt-audit

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈137 tokens for metadata: the name and description. ≈2.5k when used: this file. ≈6.6k more on demand in 2 files.

Description uses 6.9% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈548 characters, or 6.9%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Bolt CMS Project Audit (bolt-audit)

Checks an existing Bolt CMS 6 project against all skills in this repository and reports what to improve. The audit changes nothing in the project: no scaffold, no installer, no edits, no cache clear. Fixing comes afterwards, through the individual skills, once the user has read the report.

The skills are applied in WORKFLOWS.md phase order; the audit checks them in the same order. Files in this skill:

File Use
scripts/audit.sh Profile + every skill's verify.sh with full logs + example comparison + summary
references/example-map.tsv Where each skill example lives in a project, and whether drift matters (logic) or is expected (design)
references/review-checklist.md Per skill: what the script covers, what to review by hand, severity hints, defect vs. deviation
references/report-template.md Structure and severity scale of AUDIT.md

Rules

  1. Read-only. Never run a skill's scaffold.sh, privacy-config.php, merge-xliff.php, composer install/update/require, npm install, bin/console cache:clear or any write command in the audited project, and never edit its files — even to "quickly fix" something. Exceptions, all runtime data: the report directory; var/cache/<env> warmed by the verify scripts' bin/console calls; var/cache/ecs (ECS keeps its cache there, as make cscheck does). The Tailwind build check compiles into a temporary file and the timezone check boots the kernel with temporary cache folders, so neither touches the theme or var/cache/prod. Live checks are GET requests, but against a running site the asset delivery check (bolt-asset-caching) makes Bolt generate and save the thumbnails it requests — the same as a visitor's first view.
  2. Evidence for every finding: a file and line, a log line or an HTTP response. If you could not check something, list it under Not checked — do not guess.
  3. Defect or deviation. A project built differently from the skills' conventions is not broken. Report a deviation once per skill with the cost of aligning; report defects individually (see the top of review-checklist.md).
  4. Live checks only against the right site. Use --base-url only with a URL the user confirmed belongs to this project. For bolt-htaccess it must be served by Apache (staging), not symfony serve.
  5. Operator questions stay questions. API consumers, the Consent Mode decision, legal facts of the privacy policy — ask, or list them under Not checked.

Step 1 — Prepare

  • Locate the skills: SKILLS=vendor/tomvondracek/bolt-skills/skills in a project that has the package, otherwise the path of a checkout of this repository. The audit does not require the package to be installed in the audited project.
  • Ask the user (one message):
    • the URL of a running instance of this project for live checks (local, staging; none is fine);
    • whether anything consumes the REST/GraphQL API;
    • how the site is deployed (SFTP upload, git pull, CI) — it decides how bolt-security-hardening reads the committed .env;
    • which skills do not apply on purpose (for example a site that deliberately has no theme build).
  • Note the git state: the audit reports the commit it saw (profile.md).

Step 2 — Run the automated part

From the project root:

bash $SKILLS/bolt-audit/scripts/audit.sh --base-url=https://staging.example.cz   # or without --base-url
#   --project=/path/to/project   audit another directory
#   --delivery-url=https://…      site for the asset delivery check (default: --base-url when
#                                 Apache / Nginx / LiteSpeed answers there; never symfony serve)
#   --out=DIR                     report directory (default var/bolt-audit/<timestamp>/, kept out of git
#                                 through the clone's local .git/info/exclude)
#   --skip=bolt-htaccess          skills that do not apply on purpose
#   --only=bolt-cookie-consent    re-run selected skills
#   --strict                      exit 1 when a skill fails (CI)

It writes profile.md, summary.md / summary.tsv, logs/<skill>.log, examples.md and diffs/. With --base-url it also requests every sitemap URL (with each secondary-locale prefix), the search page and a missing URL, and lists the status codes in http.md (http-smoke row: any 5xx is a failure, any other non-200 a warning, the missing URL must answer 404) — pages that crash are invisible to the file-based checks. The asset delivery check of bolt-asset-caching (where each thumbnail and upload comes from, ETag, 304s) runs against --delivery-url, or against --base-url when its Server header names Apache, Nginx or LiteSpeed; profile.md says which (Asset delivery check). Against symfony serve it is off — pass --delivery-url with a staging or production URL, or start scripts/serve-apache.sh in the project first. Applicability is decided automatically where it can be: bolt-htaccess is N/A on Nginx-only sites, bolt-add-language runs once per secondary locale and is N/A on single-language sites. bolt-cookie-consent and bolt-legal-cz run without --draft: placeholders count as failures in an audit. For the register comparison run bash $SKILLS/bolt-legal-cz/scripts/verify.sh --ares once (it needs internet access).

Step 3 — Read the profile

profile.md decides how to read everything else:

  • Bolt version. The skills target Bolt CMS 6. On Bolt 5 or older, say so at the top of the report; findings that depend on Bolt 6 behaviour become "after the upgrade".
  • Theme generation. No components/, containers/, pages/ → the theme predates the convention: expect bolt-theme-init / bolt-theme-templates failures that are deviations.
  • Web server, Docker, locales, tracking found, privacy: block — they decide which checklist points apply.

Step 4 — Review the skills one by one

Go through the skills in summary.md order. For each skill:

  1. Read logs/<skill>.log completely.
  2. Read the skill's SKILL.md — its rules, pitfalls and the parts of references/ it names.
  3. Open its rows in examples.md. For logic files read every diff: a removed rule is a finding, a deliberate improvement is not. For design files check that the contract survived (data hooks, includes, required attributes), not the markup.
  4. Work through its section in references/review-checklist.md: interpret the script result (defect or deviation?), then check the Review points in the project's files.
  5. Write down the findings with evidence, severity and fix, and the skill's one-line assessment: compliant, defects, deviation (convention not used) or not applicable.

Do not skip a skill because its script passed — the Review points are exactly what scripts cannot see. Keep the notes per skill; a long audit may be continued in a new session from the report directory. Large projects: the skills are independent at this step, so the review can be split between parallel agents (one or a few skills each) if the user asks for it; merge their findings in Step 5.

Then read http.md (every non-200 page is a finding with the template that renders it) and check the Beyond the skills list at the end of the checklist.

Step 5 — Write AUDIT.md

Follow references/report-template.md: summary with the three things to fix first, a severity count, one row per skill, findings ordered Critical → Info with evidence and fix, deviations, Not checked, next steps in WORKFLOWS.md phase order. Save it in the report directory next to summary.md.

Step 6 — Hand over

Give the user the path of AUDIT.md, the severity count and the Fix first list in a few lines. Offer to fix findings — each through its skill (e.g. "run bolt-disable-api § 1–5"), one skill at a time with a commit after each, and re-run audit.sh --only=<skill> to confirm. Do not start fixing without the user's go-ahead.


Re-auditing and CI

  • After fixes, audit.sh --only=a,b re-runs just those skills into a new report directory; compare summary.tsv files.
  • In CI, audit.sh --strict --out=audit-report fails the job on any failing skill; store the directory as an artifact. Without a reachable site the live checks are skipped, not failed.

Pitfalls

  • Reading failures literally. Many failures on an older project only say "not built with this convention". Decide defect vs. deviation per the checklist before rating.
  • Live checks against the wrong server. Without --base-url the script points live checks at a closed port on purpose; never rely on a default URL.
  • .htaccess rules on PHP's built-in server. They are not applied there — bolt-htaccess live results only count against Apache. Locally, bolt-asset-caching/scripts/serve-apache.sh start serves the project with its own .htaccess files; the htaccess and delivery checks send X-Forwarded-Proto: https to it so the root HTTPS redirect stays quiet.
  • Local live checks read .env.local. A developer's .env.local (APP_ENV=dev, APP_DEBUG=1, empty SYMFONY_TRUSTED_HOSTS / BOLT_CANONICAL) makes a local instance fail the security live checks (debug token, stack traces, foreign Host) and send X-Robots-Tag: noindex (Symfony's disallow_search_engine_index follows kernel.debug). Those are not findings about the committed .env. To see production behaviour locally without editing files, start the server with the committed values as real environment variables — they win over .env.local — e.g. APP_DEBUG=0 SYMFONY_TRUSTED_HOSTS='…' BOLT_CANONICAL=https://… bash serve-apache.sh start and send Host: <domain>; otherwise report those live checks against production only.
  • Missing vendor/. Several checks need composer install; the profile says so. Ask the user to install dependencies (in their environment) rather than doing it yourself.
  • The problem/warning line counts in summary.md are a size indicator only — the logs are the source.

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No third-party reports yet.

Signed by skilld at 083eb14. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 3 days ago.

Activeupdated 2 weeks ago

README badge

README badge for vondry/bolt-skills/bolt-audit