HandL AI Connector Access Control
| 开发者 | haktansuren |
|---|---|
| 更新时间 | 2026年8月21日 12:15 |
| PHP版本: | 7.4 及以上 |
| WordPress版本: | 7.0 |
| 版权: | GPLv2 or later |
| 版权网址: | 版权信息 |
详情介绍:
Know when your WordPress plugins use AI.
See activity. Control access. Track estimated spend.
Avoid surprise AI bills
- See estimated spend by plugin.
- Add your own provider rates.
- Estimates are not bills.
- This plugin does not sell AI usage or add AI charges.
- Your AI provider controls actual billing.
- Shows which plugin made each AI Client request.
- Shows whether each request was allowed or blocked.
- Shows the AI task, provider, model, tokens, and estimated spend.
- Lets you allow or deny each plugin.
- Lets you control text, image, speech, and video separately.
- Includes Learn mode so you can watch first.
- Includes Emergency stop for urgent shutdowns.
- WordPress site owners.
- WordPress administrators.
- Multisite teams that manage rules one site at a time.
- Install the plugin.
- Open Settings → HandL AI Connector Access Control.
- Turn on Learn mode.
- Review Activity.
- Add rules after you know which plugins need AI.
- Plugin rules: Allow, deny, or use the site default.
- Temporary allow: Optional expiry on Allow rules (auto-reverts to the site default).
- AI type rules: Control text, image, speech, and video separately.
- Learn mode: Record activity without enforcing deny rules.
- Emergency stop: Block AI Client calls across the site.
- Blocked AI tools: Stop prompts that offer blocked tools.
- Role gate: Choose which WordPress roles may start AI operations.
- Denial alerts: Get an optional email when a request is blocked.
- Weekly report: Review activity, denials, and estimated spend by email.
- Estimated spend: Use token counts and your provider rates.
- Plugin AI profile: Open a plugin from Activity or Dashboard estimated spend to see its usage, incidents, estimated spend alerts, and current rules in one read-only view. Rule changes and CSV exports stay on their existing screens.
- Rules transfer: Export or import rules as JSON.
- Activity limits: Control how many entries are saved and for how long.
- Audit report: Print a governance summary (rules + activity) as PDF from your browser.
- Multisite overview: View status and activity for each site.
- WP-CLI: List and update AI type rules from the command line.
- AI Client calls are allowed by default until you change a rule.
- This plugin governs calls made through the WordPress AI Client.
- Version 1.2.0 can observe some direct AI connections. It does not block them.
- The known-provider list is not complete.
- Caller identification is best effort.
- Experimental model force is not a spending guarantee.
- Activity may be empty when WordPress disables AI site-wide.
- Activity stays on your WordPress site by default.
- Logging, denial alerts, and webhooks are optional.
- The weekly report is selected by default.
- It sends only while logging or Learn mode is on.
安装:
- Upload the plugin folder to
/wp-content/plugins/handl-ai-connector-access-control/ - Activate the plugin through the Plugins screen in WordPress
- Go to Settings → HandL AI Connector Access Control to configure plugin rules
屏幕截图:
常见问题:
How is this different from the WordPress AI plugin’s Connector Approvals?
Connector Approvals in the WordPress AI plugin controls which plugins and themes can use configured AI connector credentials. HandL AI Connector Access Control governs AI Client prompts through wp_ai_client_prevent_prompt, adding per-plugin allow/deny rules, a capability-family matrix, tool-arming denial, shadow-AI detection for direct connections outside the AI Client, and estimated spend / denial alerting. Both can run together because they govern different layers. See Dashboard → “Beyond Connector Approvals”.
Does this stop all AI usage?
Only AI calls made through the WordPress AI Client APIs that pass through wp_ai_client_prevent_prompt. The shadow-AI detector can observe direct HTTP to known AI hosts (when logging or Learn mode is on) and, if you enable Block direct calls to known AI providers, can block those requests with optional per-plugin exceptions. AI Client traffic is never treated as shadow traffic.
What does “outside the AI Client” mean on Audit & log?
A plugin (or other PHP code) issued a WordPress HTTP request to a known AI provider host without going through the AI Client path this plugin gates. With blocking off, those rows are labeled observe. With Block direct calls to known AI providers on, they may be denied or allowed as an exception. Plugin Allow/Deny rules still apply only to AI Client traffic.
Is attribution perfect?
No. It is best-effort and may be unknown or ambiguous for some execution paths (cron, REST bootstraps, shared libraries, MU plugins). Experimental model force uses the same attribution: a pin follows the detected caller, not a guarantee of which product “owns” the spend.
Does experimental model force guarantee cost control per plugin?
No. It pins the route for calls we attribute to that plugin’s nearest stack frame. Unattributed calls are unforced by default (configurable). Misattribution can apply another plugin’s pin without failing closed — existence of a resolved plugin is not proof it is the right one.
Can I manage rules with WP-CLI?
Yes. With WP-CLI available and this plugin active:
wp aicac rule list — table of every installed plugin (active and inactive) with family-level allow/deny/inherit state. Use --format=json for machine-readable output.
wp aicac rule set <plugin-basename> <family> <allow|deny|inherit> — set one capability-family cell (text, image, speech, tts, video). Writes through the same sanitized policy path as the Rules tab. Plugin basename must match an installed plugin file (e.g. acme-plugin/acme-plugin.php), including inactive plugins.
更新日志:
Unreleased
1.6.0
- In versions 1.3.0–1.5.0, changes made on the Rules tab did not save. Version 1.6.0 saves Rules again and prevents an incomplete save from erasing rules that were not submitted.
- Change history on the Rules tab records who changed rules or settings, when, and a secret-safe before-and-after summary. The recent trail is kept separately from Activity, always records Emergency stop changes, and appears in printable audit reports.
- Dashboard governance setup score (0–100): a weighted checklist of local settings and saved Activity, with Complete, Needs setup, and Not applicable states linked to the relevant screen. Setup guidance only, not a security or safety rating.
- Optional Rule notes on plugin rules, shown on Rules, plugin profiles, and applicable Activity entries. Notes are stored locally, included in rules exports, Activity CSV exports, and printable evidence, and preserved on historical Activity entries when a rule changes or is removed.
- WP-CLI policy apply: preview a secret-safe policy diff with --dry-run, then apply the reviewed JSON export with --yes. A site mismatch requires explicit confirmation, and the previous policy is kept as a restore snapshot.
- Starter policy packs on the Rules tab: preview Strict, Balanced, or Observe-first, download a JSON backup, then apply without silently replacing your existing per-plugin rules.
- Activity keep period on the Activity tab (30, 90, 180, or 365 days, or forever): daily cleanup in batches, an optional CSV download before the first cleanup, and a Site Health line showing the period and last cleanup time.
- Optional weekly rules backup email on the Rules tab, off by default: sends a JSON rules export once a week, stores only the latest backup on the site, and lets you download or compare it.
- Insights daily trend lines for calls, estimated spend, and blocked calls over up to 30 days of saved Activity, with honest short-window labels and expandable per-plugin trends.
- One email when a plugin that never used AI makes its first AI Client call. Snooze and Quiet hours still apply. This does not change Allow or Deny.
- Share-safe policy export via WP-CLI:
wp handl-aicac policy export --redactedreplaces alert email and webhook URLs with present/absent placeholders, strips rule notes, and omits the site URL. Regular export is unchanged. Import of a redacted file keeps live secrets. - Emergency break-glass via WP-CLI: temporarily allow all AI Client calls for a set number of minutes, then restore the previous policy automatically. Cancel early to restore now. Calls during the window are tagged in Activity.
- Blocked calls now record the page context and the error returned to the calling plugin, so Activity shows what happened instead of only showing that the call was denied.
- Shared email layout for HandL alerts and reports: same header, intro, and footer on every message, with a plain-text and HTML part. Per-message wording inside the content block is unchanged.
- Optional weekly governance digest email: last 7 days of AI Client calls, estimated spend vs the previous 7 days, blocked calls, direct connections, usage spike alerts, and top plugins. Off by default.
- Per-plugin estimated monthly budgets on the Rules tab, with progress, a Dashboard banner when a budget is reached, a budget-hit email, and a Site Health notice. Choose Block or Observe-only mode when the estimated spend reaches the ceiling. Amounts are estimates, not bills.
- Optional provider or model change alerts: send one email and optional webhook the first time a plugin uses a new provider. You can also alert on new models. The first recorded call sets a baseline and does not send an alert. Alerts do not change Allow or Deny rules.
- WP-CLI read-only inspection:
wp handl-aicac policy listandwp handl-aicac log summary(table or JSON) for scripted audits without opening wp-admin. - Webhook delivery log for the last 20 attempts on Activity, a Send test webhook button next to the Webhook URL, one automatic retry after a server error or timeout, and a failure email if the retry does not succeed.
- Policy checks on the Rules tab: save sample AI calls with expected Allow or Deny results. The plugin re-checks them before rules, presets, restores, or imports are applied and warns if a result would change.
- Added optional Quiet hours on the Activity tab. Set weekly schedules to block AI Client calls or use Observe-only mode, which logs activity without blocking. The Dashboard shows when a schedule is active, and Emergency stop always takes priority.
- The Activity settings screen now shows how much space the activity log uses and estimates what shorter time limits would remove. The suggested limit only fills the field. Select Save Activity settings to apply it.
- Mute alert emails and webhooks for one plugin for 1 hour, 8 hours, 24 hours, or 7 days. Rules still apply and activity still logs. The plugin records how many alerts would have been sent during the mute.
- Compare a previous JSON rules export with your current policy on the Rules tab. See every setting that differs before you import. Unknown fields from a newer export are listed as not comparable. Nothing is changed until you use Import.
- Temporary Allow rules can send one warning email when about 24 hours remain before expiry. Renewing a rule clears the warning for that rule.
- Optional: flag newly activated plugins for review and either block or log their AI calls until you choose Allow or Deny. Existing plugins are left alone when you turn this on.
- Create a rule from an Activity call: open the Rules form pre-filled with that plugin, AI type, and provider/model. Already-covered plugins show a status instead of adding a second rule.
- Added a read-only Plugin AI profile. Click a plugin in Activity or Dashboard estimated spend to see usage by day, operation, and model; blocked calls; direct connections outside the AI Client; estimated spend alerts; and the rules that currently apply. Rule results use the same decision process as enforcement. Links open the existing Rules and CSV export screens.
- Added an optional, unchecked onboarding choice to receive product news and related offers from HandL Digital. If selected, the plugin sends the alert email address, site URL, plugin version, and consent time to HandL Digital. Privacy / Data documents the transmission and opt-out method.
- Insights: weekly call and estimated-spend trends (last 8 weeks) with per-plugin sparklines and week-over-week change. Weeks with no retained log data show as “no data kept,” not zero.
- Optional monthly email of the printable audit report, off by default. It sends a short summary and HTML attachment starting on the first of each month. If no activity was retained, it sends a no-activity note without an attachment.
- Restore a previous policy from the Rules tab. The plugin keeps the last 5 versions of your rules and settings, and shows what will change before you restore.
- Scan active plugins for possible embedded AI API keys. Possible findings appear on the Dashboard and in Site Health. The Dashboard shows only the last 4 characters, and full keys are never stored.
- Temporary Allow rules that expire on their own, with remaining/expired status and one-click renew for 7 days.
- Policy presets on the Rules tab, with a preview of every setting that will change before you apply.
- WordPress Dashboard widget for mode, Emergency stop, last-24-hour activity and estimated spend, plus a Review policy link.
- Printable audit report with current rules, activity, and estimated spend compared with alert thresholds. Save it as a PDF from your browser. Nothing leaves the site.
- Estimated month-end spend on the Dashboard and Insights after at least three days of logged activity, with a Dashboard warning and optional monthly email if the projection crosses an alert threshold.
- Dashboard and Site Health visibility when alert emails or webhooks fail repeatedly, with buttons that send tests through the existing email and webhook paths.
- One-time What’s new notice after each update, with a short in-plugin highlights panel.
- Added a first-run setup wizard. Watch activity, set up alerts, and send a test email.
- Added a policy simulator. Preview how unsaved rules would handle a sample AI call.
- Added read-only REST API endpoints for policy, activity summaries, and Site Health.
- Added optional blocking for direct calls to known AI providers. It is off by default.
- Added optional usage spike alerts. They compare today's call volume and estimated spend with the recent daily average.
- Added optional estimated-spend alerts for site-wide and per-plugin thresholds.
- Added optional email alerts for direct AI provider connections. Choose immediate alerts or an hourly summary.
- Added test email buttons for blocked-call alerts and weekly reports.
- Added CSV exports for the activity log and rules.
- Added a Site Health check for AI access control.
- Improved bulk-action and Emergency stop warnings.
- Added clear guidance about how HandL differs from Connector Approvals in WordPress AI.
- Rewrote the WordPress.org description with shorter sentences and clearer estimated-spend language. Estimates are not bills.
- Strengthened permission and security checks for admin actions.
- Fixed plus-addressed emails such as user+tag@example.com in test-email confirmations.
- Rewrote the WordPress.org listing in plain language so site owners can quickly understand what the plugin does.
- Rewrote admin screens and emails in clearer, shorter language.
- Added optional rates for each AI provider when calculating estimated spend. Estimates are not bills.
- Added an optional maximum age for Activity entries, alongside the entry limit.
- Select multiple plugins on the Rules tab and set them to Allow or Deny together.
- Limit AI Client access by WordPress role. Off by default, so every role is allowed.
- On multisite, view sites where the plugin is active, check their status, and open each site's Activity screen. View only; no network-wide rule changes.
- Send optional blocked-call alerts to a webhook. The payload follows the same privacy limits as email alerts, and you can send a test.
- Use WP-CLI to list and set per-plugin AI type rules.
- Export and import Rules as JSON with a preview and confirmation. The Activity log is not included.
- F7: Weekly report email — Dashboard aggregates via wp_mail on a weekly cron (coverage, denials, estimated spend, pins).
- Aggregates only: no prompt text, user names, or request paths leave the site in the report.
- Selected by default; reports sent only while logging or learn mode is on; Settings toggle + footer link in every email.
- Self-dating window from retained log timestamps (WP-cron lateness acceptable for a summary).
- Privacy / Data documents the weekly report exit surface (ships with this feature).
- Screenshot captions updated for the current Dashboard / Activity / Insights / Exceptions UI.
- A11y: aria-describedby on kill-switch Exceptions group (#16).
- Rules: kill-switch Exceptions is a scrollable checkbox list (no Cmd/Ctrl multi-select). Name and plugin path on separate lines.
- Exceptions are dimmed while the kill switch is off, with a note. Exception selections still save if kill is off.
- Copy diet: label “Exceptions”; one line — excepted plugins still follow normal allow/deny and capability-family rules.
- F5: Dashboard-first admin (Dashboard / Rules / Activity / Insights). Coverage tile measures known AI activity (through AI Client vs outside — not governed by these rules) with log window, span, and saturation notice.
- F5: Pin-hold tile (X of Y attempted forces) plus separate unattributed never-evaluated count; learn-mode pin_matched observation (not would-succeed).
- F5: Block-that-one from Dashboard (single-click deny + undo). Shadow callers listed as not governable.
- F5: Settings demoted under Rules collapsible panel. Cleanup: unforced count branch, has_any_force_rules bool check. F6 known limits: first-of-N on user, keep observation ts on flush, _n for first-of-N.
- Shadow-AI detector (observe only): log plugins that call known AI provider hosts over WordPress HTTP while bypassing the AI Client.
- Curated host list (OpenAI, Anthropic, Google Generative Language, Cohere, Mistral, Groq, Together, Fireworks, Perplexity, xAI, DeepSeek, OpenRouter, …); not a complete inventory.
- Excludes core AI Client / php-ai-client stack traffic (those paths already hit
wp_ai_client_prevent_prompt). - Retained rows:
channel=direct_http, host, path-only (no query), shadow provider id, decisionobserve— no body, no Authorization headers. - STORAGE: direct_http rows share the same ring buffer as AI Client rows (one window so coverage buckets can sum).
- Chatty-host collapse: same attributed plugin+host (or unattributed file+host) with ~5 min idle timeout adds incoming call tallies into one row; active clusters move to the tail; first file/caller/path kept;
count= HTTP calls (not page loads) via in-request tally + shutdown flush. - Same log gate as other observability (log_enabled or learn mode).
- Audit filter “Outside AI Client”; Insights one-line sum of call counts only (does not invent a second coverage %; sums collapsed
counts). - Privacy / Data section documents host + path-only retention for direct-HTTP observations (ships with the PR that starts retaining the field).
- Lisa gate-1 follow-up: count unit = calls; collapse moves to tail; unattributed keys include file; first observation fields retained.
- EXPERIMENTAL: per-plugin force of AI Client generations to a provider/model (Rules table columns; empty = off; labeled EXPERIMENTAL in the UI).
- Unattributed calls: explicit control (default don’t force; optional force to an admin-picked provider/model — not a site-wide pin for attributed plugins).
- Pins follow detected caller (best-effort nearest plugin frame) — not a spend guarantee; readme/UI state the bound.
- Unforced-count surface from retained audit rows when pins exist and callers resolve unknown.
- Guardrails: clone-compat cheap pre-check (not the safety); final-route verification with exact provider matching (no substring); fail-closed (throw → WP_Error) on mismatch; health option writes only on status transition; persistent admin health warning when unhealthy; does not change allow/deny.
- Uninstall removes model-force health option.
- Does not send data off-site (force is local routing preference only).
- Denial email alerts (opt-in): immediate rate-limited mail or hourly digest via wp_mail; attributed to HandL AICAC.
- Privacy section documents what alerts transmit, that they are opt-in/off by default, and that mail uses the site's transport.
- Alert mail uses path-only URI (query string stripped); full URI remains local-log only.
- wp_mail failures are contained (throwing SMTP replacements cannot fatal a denied AI call).
- Alert send deferred to shutdown so the AI Client denial filter path does not block on SMTP (connection release is not claimed — FastCGI typically holds until shutdown completes).
- Immediate-mode failures and rate-limit overflow drain via the same hourly cron safety net (scheduled whenever alert_on_deny is on, not only in digest mode).
- Digest queue cleared when alerts are disabled; uninstall removes digest queue, rate option, and cron event.
- Digest cron self-heals on init if the scheduled event was lost.
- Honesty banner when AI is disabled site-wide via wp_supports_ai (audit intentionally empty because core short-circuits before our filter).
- Estimated-$ column on the recent-call log from retained token counts × configurable $/1M rates (observability only; labeled est./default rates).
- Inferred provider/model values labeled "inferred" in the audit UI (observability honesty — not enforcement).
- PHP 7.4 nested-POST filter_input path for the operation matrix verified PASS on PHP 7.4.33 cgi.
- AI tool arming (caller intent): deny prompts that arm denied tools (
functionDeclarations— WordPress abilities and custom tools) at prevent time. - Denied tools list on Rules tab; registered abilities shown as a helper subset (not an enumeration of everything matchable).
- Case-insensitive matching; entries that match no currently registered ability are flagged (still saved — pre-listing is allowed).
- Loud denials: log
denial_reasonandmatched_toolson every denial (including non-tool denials that also armed a blocked tool). - Snapshot extracts
armed_toolsfrom model config without reflection. - Option key
denied_tools(migrates legacydenied_abilitieson read).
- Per-plugin × capability-family operation matrix (Text / Image / Speech / TTS / Video) on the Rules tab.
is_supported_for_*and matchinggenerate_*/convert_text_to_speech*map to the same family rule.- Generic
is_supported/generate_resultresolve family via core capability inference (no silent unknown bypass). - TTS prefix heuristics run before Text so unmapped TTS method names cannot misclassify as Text.
- Snapshot is taken before the allow/deny decision so the operation is known at prevent time.
- Explicit unknown-operation fallback (inherit plugin rule, allow, or deny).
- Kill-switch exceptions fall through to normal plugin + family rules (do not widen access).
- Recent-call log records and displays capability family.
- Suggested-rules column renamed to "Plugin-level would enforce" (honest: not full matrix).
- Recent-call log records input (prompt) and output (completion) token counts via
wp_ai_client_after_generate_resultwhen logging is enabled. - Usage insights tab: aggregated call counts and token sums/peaks by plugin, provider, model, and operation from the retained log.
- Learn mode (Activity tab): log every AI Client call without blocking; suggested rules from the log; “would enforce” decision in the audit trail.
- Emergency kill switch: block all AI Client calls with optional per-plugin exceptions.
- Quick actions on the recent-call log: allow or deny a plugin in one click.
- Added WordPress.org plugin directory banners and screenshots.
- Enriched recent-call log with provider, model, operation, and truncated prompt preview; improved admin log table and documented count-based retention (no TTL).
- Added logos + branding
- Renamed plugin to HandL AI Connector Access Control (slug
handl-ai-connector-access-control). Prefixed options, constants, menus, and forms per WordPress.org guidelines. Migrates settings from previous plugin slugs when present.
- Initial release (submitted as AI Not).