AdAnt v0.11.0 — server research workflow

What changed

• Codex and Claude now research through the same server workflow as AdAnt Studio: profile the product, collect social evidence, analyze selected videos, and save the report.
• Normal research no longer requires a local Python runtime, browser recovery, checkpoint files, PDF rendering, or an upload sequence. Reports are saved directly to Studio with on-demand PDF export.
• Queued profile, seed-metadata, and analysis jobs can continue while the host is disconnected. Resume uses the same run and operation IDs; curation and writing continue when the host resumes.
• Both hosts share the web workflow's seed exclusions, evidence checks, platform scope, attempt limits, and explicit partial-report behavior.

Canonical AdAnt source: b5e64659a706ac95b3a7e1e0fda9c65b6db77303
Public release commit: d47397c0f059dfbe756bd14851f4f41986270cb8

Validation

• Plugin suite: 61 tests passed, no skips.
• Codex plugin validator passed.
• Claude plugin and marketplace strict validators passed.
• v0.11.0 release-sync reproducibility check passed.
• Public diff and credential/private-content scan reviewed; no credentials or private service URLs found. Two existing dependency-maintenance comments were reviewed as benign matches.
• Claude release-tag dry run confirmed adant--v0.11.0.

Install or update

For a new Codex CLI installation:

codex plugin marketplace add anyloop/agent-plugin --ref main
codex plugin add adant@adant-ai

To update an existing Codex CLI installation:

codex plugin marketplace upgrade adant-ai
codex plugin add adant@adant-ai

For a new Claude Code installation:

claude plugin marketplace add https://github.com/anyloop/agent-plugin.git
claude plugin install adant@adant-ai

To update an existing Claude Code installation:

claude plugin marketplace update adant-ai
claude plugin update adant@adant-ai

In the desktop app, update AdAnt through Plugins. Start a new task or conversation after updating so the new skill instructions are loaded; complete any connection prompt shown by your host.

This release requires the new AdAnt research tools on the server. Missing tools are reported as a capability gap. Live end-to-end performance has not been benchmarked; this release does not promise a specific completion time.

AdAnt 0.10.3 — authentication preflight and hashtags

What changed

• Verify the remote AdAnt connection before authenticated work in both Codex and Claude Code. Failed or unverified authentication now stops early with the observed error and the correct host-specific recovery step.
• Verify local credentials instead of treating a stored token file as proof of authentication. Permission, service, and network failures remain distinct and actionable.
• Use AdAnt's server-side social suppliers without asking users to sign in to TikTok or Instagram. Public browser discovery remains an evidence-gap fallback after authentication succeeds.
• Add product-specific hashtag recommendations to initial social research and include them in research reports and strategy briefs.

Canonical AdAnt source: b929ec1121c69c1c09742bd552c133cd7a16ce0b

Validation

• AdAnt plugin suite: 61 checks passed.
• Codex plugin validator passed.
• Claude plugin and marketplace strict validators passed.
• Release sync reproducibility check passed for v0.10.3.
• Public-package credential and private-infrastructure scan found no matches.

Update

Codex CLI:

codex plugin marketplace upgrade adant-ai
codex plugin add adant@adant-ai

Claude Code:

claude plugin marketplace update adant-ai
claude plugin update adant@adant-ai

Start a new task or conversation after updating so the new skill instructions are loaded.

AdAnt 0.10.2 — Python certificate fix

Fixes social research failing with CERTIFICATE_VERIFY_FAILED on Python installations without a default system CA certificate file.

Research subprocesses now receive the bundled certifi CA file when needed. Explicit certificate settings are preserved, and HTTPS certificate and hostname verification remain enabled. The fix ships in both the Codex and Claude Code plugins.

Validation

• All 61 plugin checks passed, including the local-server regression suite.
• Codex plugin validation passed.
• Claude plugin and marketplace strict validation passed.
• Public synchronization check and diff review passed.

Canonical AdAnt source: a4304184c239d1c27dd5a9fd7379f8cf8cc2d7f0 (read-only; no AdAnt changes or publication performed by this release).

Install or update in Codex

For an existing Git marketplace:

codex plugin marketplace upgrade adant-ai
codex plugin add adant@adant-ai
codex plugin list | grep -i adant

For a first installation:

codex plugin marketplace add https://github.com/anyloop/agent-plugin.git --ref main
codex plugin add adant@adant-ai

If the marketplace is registered as a local directory, update that checkout or replace its registration with the Git URL before using marketplace upgrade.

Verify version 0.10.2, then fully quit and reopen Codex to replace any running server from the previous installation. Return to your task and retry the research request.

Claude Code

/plugin marketplace update adant-ai
/plugin update adant@adant-ai

Restart Claude Code after updating to load the new local runtime.

AdAnt 0.10.1 — reliable setup and strategy briefs

AdAnt 0.10.1 improves first-run reliability and the social strategy workflow.

What changed

• Added an explicit local runtime setup command that installs uv without changing shell profiles and prepares locked dependencies before the MCP connection.
• Improved initialization diagnostics so remote authentication and local startup failures are handled separately, with MCP reconnect preferred over restarting the desktop app.
• Initialization now continues an existing request after readiness checks instead of replacing it with another starter menu.
• Added focused social strategy briefs with fresh platform evidence, prior-video exclusion, PDF delivery, and saved AdAnt reports with Create in Studio actions.
• Improved strategy keyword mining from example videos and existing reports, and documented server-side social discovery as the primary path.

Canonical AdAnt source: 4907db710610eda901103cc275d42fdd7af73d54

Validation

• Canonical plugin suite: 61 tests passed with no skips.
• Codex plugin manifest validation passed.
• Claude plugin and marketplace strict validation passed.
• Generated public repository matched the canonical sync output for version 0.10.1.
• Public diff was reviewed for secrets, private endpoints, credentials, customer data, and deployment details.

Install or update

Codex CLI:

codex plugin marketplace add anyloop/agent-plugin --ref main
codex plugin add adant@adant-ai

Claude Code:

claude plugin marketplace add https://github.com/anyloop/agent-plugin.git
claude plugin install adant@adant-ai

AdAnt 0.10.0 — improved social research

Changes

• Search social platforms through AdAnt first, using browser sessions to fill evidence gaps, with a dedicated progress stage.
• Improve abbreviated engagement counts and keep Instagram likes separate from views.
• Balance product relevance with engagement evidence, preserve audited metrics and durations, and identify weaker context examples explicitly.
• Provide concise copy-ready strategy briefs naming the source video, product, and avatar.
• Use one local device identity for authenticated research requests and simplify setup checks.
• Pin the video downloader for consistent acquisition behavior.

Source and validation

Canonical AdAnt source: c9a40916ec5aa6eec8a18e27a987dcfa754d459d.
The canonical repository was not edited, committed, tagged, or pushed by this release workflow. Uncommitted local plugin fixes are excluded.

Passed: widget reproducibility check; all 56 plugin contract/runtime checks; 41 local-server tests; 26 report tests; strict Claude plugin and marketplace validation; Codex plugin validation; synchronization check. Version 0.10.0 passed the final manifest and synchronization checks.

Install

Codex:

codex plugin marketplace add anyloop/agent-plugin --ref main
codex plugin add adant@adant-ai

Claude Code:

claude plugin marketplace add https://github.com/anyloop/agent-plugin.git
claude plugin install adant@adant-ai
claude mcp login plugin:adant:adant

Update

Codex:

codex plugin marketplace upgrade adant-ai
codex plugin add adant@adant-ai

Claude Code:

claude plugin marketplace update adant-ai
claude plugin update adant@adant-ai

Start a new task/conversation after updating, then invoke $adant-init.

AdAnt 0.9.0 — plugin v2

AdAnt 0.9.0 completes the public plugin v2 migration for Codex and Claude Code.

Changes

• Makes the v2 MCP workflow the only packaged execution path and removes the legacy v1 runtime, executable AnyT notebooks, and CLI fallback.
• Replaces large workflow skills with thinner MCP-only orchestration documents.
• Adds the local adant-local MCP server for durable research workflows, resumable phases, reports, local media handling, and live progress.
• Adds shared MCP App experiences for research progress and media previews.
• Adds short-lived scoped local authentication and persistent device identity for stronger session security and revocation.
• Removes the system Python and Node.js prerequisites: uv provisions the locked Python runtime automatically. Local browser research still requires uv, Chrome, and platform login where applicable.

Canonical AdAnt source: ecec1cb72ebe7c9cf844c7be2dc33019bf45d557

Validation

• Canonical plugin suite: 51 tests passed.
• Canonical monorepo gate: 4,971 unit tests passed, 1 skipped.
• Claude plugin and marketplace strict validation passed.
• Official Codex plugin validation passed.
• Exact public synchronization check for 0.9.0 passed.
• Complete public diff and private-material scan passed.

Install

Codex CLI

codex plugin marketplace add anyloop/agent-plugin --ref main
codex plugin add adant@adant-ai

Claude Code

claude plugin marketplace add https://github.com/anyloop/agent-plugin.git
claude plugin install adant@adant-ai
claude mcp login plugin:adant:adant

Update

Codex CLI

codex plugin marketplace upgrade adant-ai
codex plugin add adant@adant-ai

Claude Code

claude plugin marketplace update adant-ai
claude plugin update adant@adant-ai

Start a new task or conversation after updating.

AdAnt 0.8.3

AdAnt 0.8.3 prevents long local social-research runs from exhausting browser memory.

Changes

• Runs TikTok, Instagram, Meta Ads, and YouTube browser capture one platform at a time.
• Closes workflow-owned tabs immediately after each platform completes or fails instead of waiting for the agent turn to end.
• Fixes YouTube fallback Chrome cleanup and hardens Meta cleanup with terminate, wait, and forced-kill handling.
• Cleans up partially started TikTok and Instagram research browsers.
• Adds browser lifecycle regression coverage for the Codex and Claude Code packages.

Canonical AdAnt source: 683e0369a3901067f1f28fe8e140426754ab3c57

Validation

• Canonical plugin suite: 43 tests passed.
• Claude plugin and marketplace strict validation passed.
• Codex plugin validation passed.
• Exact public synchronization check for 0.8.3 passed.
• Complete public diff and private-material scan passed.

Install

Codex CLI

codex plugin marketplace add anyloop/agent-plugin --ref main
codex plugin add adant@adant-ai

Claude Code

claude plugin marketplace add https://github.com/anyloop/agent-plugin.git
claude plugin install adant@adant-ai
claude mcp login plugin:adant:adant

Update

Codex CLI

codex plugin marketplace upgrade adant-ai
codex plugin add adant@adant-ai

Claude Code

claude plugin marketplace update adant-ai
claude plugin update adant@adant-ai

Start a new task or conversation after updating.

AdAnt Agent Plugin v0.8.2

Canonical AdAnt source: bf244e3537c2e3acc3ec9f0741c48c6050a33965

What changed

• Social research now treats documented empty-result exits as visible fallback warnings while preserving real failures.
• Video acquisition ignores user-level yt-dlp configuration, preventing unsupported local impersonation settings from breaking selected Shorts.
• Ranked strategy analysis runs through a bounded two-slot queue with immediate reserve promotion and valid-output caching.
• Social research documents safe platform subagent ownership for independent browser contexts and shared-browser hosts.
• PDF generation waits for document, font, image decode, and layout readiness instead of paying a fixed sleep.

Validation

• pnpm check:plugins: 42 checks passed
• Strict Claude plugin and marketplace validation: passed
• Codex plugin validation: passed
• Versioned synchronization reproducibility check: passed
• Complete diff and private-material scan: passed

Install

Codex

codex plugin marketplace add anyloop/agent-plugin --ref main
codex plugin add adant@adant-ai

Claude Code

claude plugin marketplace add https://github.com/anyloop/agent-plugin.git
claude plugin install adant@adant-ai

Update

Codex

codex plugin marketplace upgrade adant-ai
codex plugin add adant@adant-ai

Claude Code

claude plugin marketplace update adant-ai
claude plugin update adant@adant-ai

Start a new Codex task or Claude Code conversation after updating.

AdAnt v0.8.1 — faster, resilient social research

Canonical source: anyloop/adant @ 85aa08259d2224a9e51919b8bd3861f6fe78d986

What's new

Bounded social-research timing. Research subprocesses now have hard phase limits, while browser-driven work uses explicit stage and end-to-end budget checkpoints. The Sidecar shows live elapsed time against the workflow target and active phase budgets, making overruns visible before they become long retry loops.

Useful results arrive sooner. Production research now publishes a readable 13-slide HTML/markdown preview immediately after curation, while the five deeper strategy analyses continue. Video understanding uses a two-job concurrency limit to avoid the transport failures observed with four-way fan-out.

More resilient video acquisition. Codex Browser sessions can pass an already-downloaded local clip while retaining its permanent platform URL. The portable yt-dlp fallback now includes browser-compatible extraction support, bounded retries, structured failure codes, successful-output-only caching, and reserve-video promotion.

Reliable report delivery. Presigned report uploads use the runtime's bundled trusted CA set, preventing first-pass certificate failures on Python installations that do not inherit the operating system trust store.

Codex and Claude remain aligned. Both public packages contain the same managed research runtime, workflow rules, acquisition behavior, report handoff, and regression coverage.

Validation

• Monorepo plugin suite: 40/40 passed, including Sidecar and acquisition Python tests.
• Claude plugin and marketplace: strict validation passed.
• Codex plugin: installed plugin-creator validation passed.
• Public sync reproducibility check: clean at version 0.8.1.
• Public diff formatting check: passed.
• Codex and Claude managed public surfaces: byte-for-byte matched.
• Complete public diff reviewed; no tokens, credentials, customer data, private/internal URLs, Brain instructions, provider credentials, or deployment details found.

Install / update

ChatGPT desktop: update the AdAnt plugin from the personal marketplace on main, fully quit and restart the desktop app, then start a new task.

Codex CLI:

codex plugin marketplace add anyloop/agent-plugin --ref main
codex plugin add adant@adant-ai

Claude Code:

claude plugin marketplace add https://github.com/anyloop/agent-plugin.git
claude plugin install adant@adant-ai
claude mcp login plugin:adant:adant

After updating, start a new task and run $adant-init.

AdAnt v0.8.0 — faster, visible social research

Canonical source: anyloop/adant @ 5b367d22cb80856f71666900f74ed13c3b9dab0b

What's new

Faster research with an explicit delivery contract. Production-complete social research now targets 30–45 minutes by running independent platform discovery and strategy analysis concurrently. A separately labeled fast diagnostic draft targets 15–25 minutes when explicitly requested, while clearly omitting deep analysis, strategy pages, top-ups, and AdAnt upload.

Useful progress throughout the run. Research now reports the current stage, findings so far, next step, ETA, and risks at bounded 45–60 second intervals. The Sidecar shows the selected workflow mode, parallel work, dynamic future stages, and intermediate results without announcing completion between phases.

Built-in Codex Browser support. TikTok, Instagram, Meta Ads, YouTube Shorts, competitor research, and social strategy workflows prefer Codex Desktop's persistent in-app Browser when available. The portable Chrome/CDP path remains available as a disclosed fallback, and competitor research can resume from an existing browser capture without repeating discovery.

Codex and Claude remain aligned. Both public packages contain the same managed research runtime, workflow plans, browser-selection guidance, data shaping, and regression coverage.

Validation

• Monorepo plugin suite: 39/39 passed, including Sidecar Python runtime tests.
• Claude plugin and marketplace: strict validation passed.
• Codex plugin: installed plugin-creator validation passed.
• Public sync reproducibility check: clean at version 0.8.0.
• Public diff formatting check: passed.
• Complete public diff reviewed; no tokens, credentials, customer data, private URLs, Brain instructions, provider credentials, or deployment details found.

Install / update

ChatGPT desktop: update the AdAnt plugin from the personal marketplace on main, fully quit and restart the desktop app, then start a new task.

Codex CLI:

codex plugin marketplace add anyloop/agent-plugin --ref main
codex plugin add adant@adant-ai

Claude Code:

claude plugin marketplace add https://github.com/anyloop/agent-plugin.git
claude plugin install adant@adant-ai
claude mcp login plugin:adant:adant

After updating, start a new task and run $adant-init.