Daily AI Implementation Scout Council

2026-08-07. Top pick: #1 code-review-graph. Each item is graded on 7 axes; copy a build command to act on it.

Today's ranked top 20

#1
newbuild nowBoth runtimes33 / 35

code-review-graph repo

MCP-native local codebase intelligence graph. Builds a persistent dependency + call graph (crg-daemon auto-rebuilds on change) and exposes four MCP tools: get_impact_radius, get_review_context, detect_changes, get_architecture_overview.

What it does for you: Gives every Claude Code review session instant knowledge of which files a change will propagate to, so review misses no downstream risk. For Hermes, any pre-commit or CI hook can call the MCP tools to gate risky diffs automatically. Trending 232 stars on Aug 7 means community validation is arriving fast.

In practice: Polished, minimal, single-purpose. Installation is pip install code-review-graph; crg-daemon then runs as a background service. 28 983 total stars at time of scouting.

For: Both runtimes. pip-installable Python daemon with 4 MCP tools; works in Claude Code and Hermes pipelines equally.

Security4
Quality5
Auditability4
Useful to you5
Useful to community5
Buildable now5
Hermes5

Verdict: build now. MCP-native, pip-installable today, fills a real gap in Hermes's code-awareness toolchain, and the 4-tool surface is small enough to integrate in one session.

Build #1 code-review-graph: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/tirth8205/code-review-graph. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · 28 983 GitHub stars, 232 gained on 2026-08-07; MCP tool list confirmed in README; pip-installable with crg-daemon auto-rebuild; Show HN thread same day.

#2
newtest firstBoth runtimes30 / 35

loopx repo

Durable goal manager for coding agents. Maintains goal state across sessions with quota-aware auto-wake, executable todos, evidence logs, and verifiable handoffs.

What it does for you: Solves the session-amnesia problem in long Axion build projects: goals persist, evidence accumulates, and the next session resumes exactly where the last stopped. Verifiable handoffs mean YY can trust a night-shift completion report without re-reading the full log.

In practice: 2 800 stars, 207 forks; Claude Code explicitly named in supported agents list. Handoff format is plain JSON so Hermes can consume it natively.

For: Both runtimes. Agent-agnostic goal-state manager; integrates with Claude Code and any Hermes loop via its JSON state protocol.

Security3
Quality4
Auditability5
Useful to you5
Useful to community4
Buildable now4
Hermes5

Verdict: test first. High strategic value for multi-session Axion builds; needs a test run to validate the handoff format integrates cleanly with existing Hermes session state.

Build #2 loopx: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/huangruiteng/loopx. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · 2 800 stars, 207 forks; Claude Code listed in supported agents; JSON handoff format inspected; quota-aware auto-wake confirmed in README.

#3
newtest firstHermes29 / 35

channels-sdk repo

Open-source SDK for building AI agent interactions inside Slack, Teams, Discord, and Telegram with native interactive UI components (buttons, forms, cards).

What it does for you: Upgrades Hermes from plain-text Telegram messages to rich interactive flows: approval buttons, quick-reply keyboards, inline cards. Solves the clunky text-only approval UX in the current bridge without touching the Telegram bot token or rewriting the bridge core.

In practice: Show HN on Aug 7; CopilotKit provenance means active maintenance. Telegram explicitly supported. Installation is npm/pip; no backend lock-in.

For: Hermes. Open-source SDK that adds structured interactive UI to Slack, Teams, Discord, and Telegram — Hermes's primary user surface.

Security4
Quality4
Auditability4
Useful to you5
Useful to community4
Buildable now4
Hermes4

Verdict: test first. Direct upgrade to Hermes UX; Telegram support confirmed. Needs a spike to verify the SDK wrapper sits cleanly over python-telegram-bot without breaking existing bridge routes.

Build #3 channels-sdk: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/CopilotKit/channels-sdk. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · Show HN 2026-08-07; CopilotKit org; Telegram listed in supported channel list in README; npm package inspected.

#4
newtest firstBoth runtimes29 / 35

open-swe repo

Open-source async software engineering agent. Follows a plan-execute-test-commit-PR-reply loop on LangGraph and Deep Agents infrastructure.

What it does for you: Gives Axion a production-grade autonomous coder that can take a GitHub issue, write the fix, run tests, and open a PR — closing the last mile of the night-shift queue. The LangGraph foundation means swap-in for any LangChain-based Hermes step is straightforward.

In practice: LangChain official repo, just released Aug 7. Deep Agents and LangGraph stack match the MCL coaching infra already evaluated. Async by default.

For: Both runtimes. Open-source async SWE agent built on LangGraph + Deep Agents; runs standalone or integrates with any Claude-based orchestrator.

Security3
Quality5
Auditability4
Useful to you4
Useful to community5
Buildable now4
Hermes4

Verdict: test first. Strong provenance (LangChain official), directly fills the autonomous-coder gap in the night-shift queue. Needs validation on the Axion Windows environment before relying on it.

Build #4 open-swe: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/langchain-ai/open-swe. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · LangChain official GitHub org; published Aug 7 2026; Deep Agents + LangGraph dependency chain confirmed; async PR loop described in README.

#5
test firstHermes28 / 35

TencentDB-Agent-Memory repo

Team-level memory hub for AI agents. Stores cross-session context and shared team knowledge in TencentDB, queryable by any agent via REST API.

What it does for you: Adds a durable, team-shareable memory layer to Hermes: client context, campaign notes, and coaching observations survive session resets and are visible to any Hermes subagent, not just the session that wrote them. Complementary to loopx (goal state) rather than duplicating it.

In practice: TencentCloud official repo; REST API is simple; TencentDB backend is replaceable. Works alongside any vector store already in Axion.

For: Hermes. Team-level persistent memory hub; integrates with Hermes via its REST API to share context across all agent sessions.

Security3
Quality4
Auditability4
Useful to you4
Useful to community4
Buildable now4
Hermes5

Verdict: test first. Fills the shared-memory gap between Hermes runs without requiring a new infra stack. REST API is a low-friction integration point. Needs validation that TencentDB credentials stay out of the Axion bridge log.

Build #5 TencentDB-Agent-Memory: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/TencentCloud/TencentDB-Agent-Memory. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · TencentCloud official org; REST API endpoint list in README; cross-session continuity and team sharing confirmed in feature list; Python client included.

#6
newtest firstBoth runtimes26 / 35

aws-agent-toolkit repo

Official AWS toolkit delivering MCP servers, agent skills, and framework plugins for S3, Lambda, DynamoDB, and other core AWS services.

What it does for you: Drops AWS resource management directly into Claude Code and Hermes without hand-rolled boto3 wrappers. S3 and DynamoDB coverage is immediately useful for the brain-backup and data-pipeline work already in Axion.

In practice: AWS official repo; MCP server pattern is the same pattern as the rest of the Axion MCP fleet. Well-documented; permissive licence.

For: Both runtimes. Official AWS MCP servers and skill plugins work in Claude Code and in any Python-based Hermes step.

Security4
Quality4
Auditability4
Useful to you3
Useful to community4
Buildable now4
Hermes3

Verdict: test first. Official provenance, low integration cost, meaningful coverage of AWS services already used. YY utility is moderate because current Axion mostly runs locally, but the backup and Drive pipelines would benefit.

Build #6 aws-agent-toolkit: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/aws/agent-toolkit-for-aws. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · AWS official GitHub org; MCP server list confirmed; S3/Lambda/DynamoDB coverage in README; permissive Apache-2.0 licence.

#7
newtest firstHermes26 / 35

AionUi repo

Always-on cowork GUI for AI coding agents. Provides MCP Unified Management, YOLO/Full-Auto Mode, and a persistent session interface. Hermes Agent is listed as a supported agent.

What it does for you: Gives YY a single desktop GUI to monitor and drive all Hermes and Claude Code sessions, replacing ad-hoc terminal windows. MCP Unified Management consolidates the growing Axion MCP fleet into one control surface. Windows-compatible confirmed.

In practice: iOfficeAI org; Hermes Agent named explicitly. Windows-compatible. MCP management panel visible in screenshots. Full-Auto mode means low-touch overnight runs.

For: Hermes. 24/7 cowork GUI that explicitly names Hermes Agent in its supported-agents list and offers MCP Unified Management.

Security3
Quality3
Auditability3
Useful to you5
Useful to community3
Buildable now4
Hermes5

Verdict: test first. Hermes name-drop is a strong signal of intended audience; MCP management fills a real Axion pain point. Needs hands-on test before committing to it as the primary cowork surface.

Build #7 AionUi: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/iOfficeAI/AionUi. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · iOfficeAI GitHub org; Hermes Agent listed in supported agents; MCP Unified Management panel in README screenshots; Windows compatibility stated.

#8
newstudyStandalone tool29 / 35

uber-ADR repo

AI agent observability, security benchmarking, and threat-detection framework from Uber Engineering. Covers runtime monitoring, adversarial testing, and audit trails for production agent deployments.

What it does for you: Provides the security audit and threat-detection layer that Axion's agent fleet currently lacks. The benchmarking suite lets YY measure Hermes's actual security posture rather than relying on code review alone. Audit trails feed directly into the friction-audit skill.

In practice: Uber official engineering repo; strong security and auditability scores. Designed for production agent fleets at enterprise scale.

For: Standalone tool. Enterprise-grade observability and security benchmarking platform; Axion would consume its reports rather than embed its code.

Security5
Quality4
Auditability5
Useful to you4
Useful to community4
Buildable now3
Hermes4

Verdict: study. Extremely high quality signal on agent security; the patterns are worth internalising for Axion's own threat model even if the full framework is too heavy to run locally right now.

Build #8 uber-ADR: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/uber/ADR. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · Uber official GitHub org; runtime monitoring + adversarial benchmarking confirmed; audit trail format documented; production scale stated in README.

#9
newstudyStandalone tool28 / 35

Deep Agents vs LangChain vs LangGraph skill

LangChain official comparison of Deep Agents, LangChain, and LangGraph covering when to use each, trade-offs, and migration paths.

What it does for you: Settles the framework-choice question for any new Axion agent build: clear decision tree for which layer to pick based on task type, durability requirement, and integration complexity. Directly informs the open-swe and channels-sdk integration decisions.

In practice: LangChain official blog, Aug 6-7 window. Authoritative primary source, not a third-party summary.

For: Standalone tool. Decision framework article; consumed as reference knowledge, not code.

Security4
Quality5
Auditability4
Useful to you4
Useful to community5
Buildable now2
Hermes4

Verdict: study. Required reading before any new LangGraph or Deep Agents integration in Axion. High conceptual value; no code to ship.

Build #9 Deep Agents vs LangChain vs LangGraph: use the ai-implementation-build-intake skill to build this safely. Source: https://www.langchain.com/blog/deep-agents-vs-langchain-vs-langgraph. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · LangChain official blog; decision matrix confirmed in article; covers all three layers; published within scouting window.

#10
newstudyClaude (Axion)28 / 35

agent-skills (addyosmani) skill

Addy Osmani's production engineering agent skills for Claude Code. Covers code review, performance analysis, test generation, and refactoring patterns from Google Chrome engineering context.

What it does for you: Battle-tested skill patterns from a senior engineering context; drop into Axion's .claude/skills/ to immediately level up Claude Code sessions. Complements the superpowers skill set already installed.

In practice: Addy Osmani (Google Chrome) provenance; production-grade patterns; MIT licence. Small focused repo.

For: Claude (Axion). Claude Code skill files; designed to drop into .claude/skills/ and be invoked directly in Claude Code sessions.

Security4
Quality4
Auditability4
Useful to you4
Useful to community4
Buildable now4
Hermes4

Verdict: study. High-quality patterns worth reviewing and selectively adopting; not a direct build but enriches the Axion skill library.

Build #10 agent-skills (addyosmani): use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/addyosmani/agent-skills. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · Addyosmani GitHub account; production engineering context stated; MIT licence; skill file format compatible with Claude Code.

#11
studyClaude (Axion)27 / 35

skills (mattpocock) skill

Matt Pocock's real-world .agents directory skills covering TypeScript, testing, and developer tooling patterns.

What it does for you: TypeScript-heavy skill patterns from a prominent DX educator; useful additions to Axion's TypeScript build steps. Different emphasis from Osmani (DX vs Chrome perf).

In practice: Total TypeScript creator; real production .agents directory; small focused repo.

For: Claude (Axion). Matt Pocock's actual .agents directory skills; Claude Code native.

Security4
Quality4
Auditability4
Useful to you4
Useful to community4
Buildable now4
Hermes3

Verdict: study. Good complementary patterns to Osmani; TypeScript focus less critical for Python-heavy Axion but valuable for any MCL web work.

Build #11 skills (mattpocock): use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/mattpocock/skills. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · mattpocock GitHub account; Total TypeScript creator; real .agents directory; skill file format inspected.

#12
studyClaude (Axion)27 / 35

superpowers (obra) skill

The canonical source of the superpowers Claude Code skills framework already running in Axion. Contains the full skill library and the using-superpowers guide.

What it does for you: Reading the source reveals unpublished skills and patterns not yet in the Axion install. Diff against the installed version to find upgrades. High signal for YY as superpowers is already the primary Axion skill runtime.

In practice: obra GitHub; the source of truth for the framework already in production. Occasionally ahead of the distributed version.

For: Claude (Axion). The Claude Code skills framework source repository; Claude Code native.

Security4
Quality4
Auditability4
Useful to you5
Useful to community4
Buildable now4
Hermes2

Verdict: study. Source-of-truth for the Axion skill runtime; worth a periodic diff to catch new skills. No build action — already installed.

Build #12 superpowers (obra): use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/obra/superpowers. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · obra GitHub; confirmed as source of superpowers skill already in Axion; skill file format matches installed version.

#13
newstudyStandalone tool24 / 35

Scalex: humans missed 1 in 3 agent threats skill

Scalex security research showing that human reviewers missed 1 in 3 agent-level permission threats across 40 000 agent interactions. Covers common threat patterns and detection gaps.

What it does for you: Hard data for calibrating Axion's agent permission model. The 33% miss rate is a concrete benchmark for how much automated threat detection Axion needs vs. how much it can rely on YY's manual review.

In practice: 40K interaction dataset; security research publication; quantitative findings.

For: Standalone tool. Security research article; informs Axion threat model design rather than embedding as code.

Security5
Quality4
Auditability3
Useful to you4
Useful to community4
Buildable now2
Hermes2

Verdict: study. Rare quantitative data on agent security gaps; directly applicable to Axion's threat model but no code to ship.

Build #13 Scalex agent threats: use the ai-implementation-build-intake skill to build this safely. Source: https://scalex.dev/blog/ai-agent-permissions-stats/. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · Scalex.dev blog; 40K interaction sample size stated; 33% miss rate finding confirmed in article body; detection gap patterns documented.

#14
newstudyStandalone tool22 / 35

Qwen3.8 Max agentic benchmark skill

Artificial Analysis agentic intelligence index. Benchmarks leading models on multi-step agent tasks; latest data shows Qwen3.8 Max near the top of the agentic ranking.

What it does for you: Live reference for which model to route Hermes tasks to. Qwen3.8 Max appearing in the top tier opens a cost-efficient alternative to Claude Sonnet 4.6 for high-volume Hermes steps that do not need MCL brand-voice quality.

In practice: Artificial Analysis is the standard independent benchmark; agentic-index is relatively new but methodologically sound.

For: Standalone tool. Benchmark dashboard; reference data for model selection decisions, not embeddable code.

Security4
Quality4
Auditability3
Useful to you3
Useful to community4
Buildable now2
Hermes2

Verdict: study. Model selection data; high value for Hermes routing decisions but no code to ship. Bookmark and revisit monthly.

Build #14 Qwen3.8 agentic benchmark: use the ai-implementation-build-intake skill to build this safely. Source: https://artificialanalysis.ai/?intelligence=agentic-index. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · Artificial Analysis published benchmark; agentic-index tab confirmed; Qwen3.8 Max position in top tier verified on scouting date.

#15
newwatchBoth runtimes23 / 35

wigolo repo

Agent red-teaming and adversarial prompt-testing framework. Runs structured attack scenarios against agent endpoints to surface permission leaks and prompt-injection vulnerabilities.

What it does for you: Provides a local adversarial test harness for Hermes before deploying new agent skills, catching permission leaks that the Scalex research shows humans routinely miss.

In practice: KnockOutEZ org; security-focused; CLI tool. Smaller repo, active development.

For: Both runtimes. Agent red-teaming and prompt-testing tool; runs as a standalone CLI and can be invoked from Hermes CI hooks.

Security5
Quality3
Auditability3
Useful to you3
Useful to community3
Buildable now3
Hermes3

Verdict: watch. Useful security tooling but smaller provenance and less polished than uber-ADR. Worth watching for maturity.

Build #15 wigolo: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/KnockOutEZ/wigolo. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · KnockOutEZ GitHub; red-team scenario list in README; CLI invocation confirmed; adversarial prompt patterns documented.

#16
newwatchBoth runtimes23 / 35

video-use repo

Agent that edits video files using natural-language coding instructions. Built on browser-use infrastructure; interprets edit commands and applies ffmpeg-style transformations.

What it does for you: Automates the video trim-and-grade step in the MCL content pipeline, potentially removing the manual ffmpeg scripting in the existing video-pipeline-tool. Most useful for the Boost Reel and SFC production flows.

In practice: browser-use org; just released; natural-language video edit interface is novel. Relies on ffmpeg under the hood.

For: Both runtimes. Video editing agent; integrates with browser-use infrastructure and runs standalone or from a Hermes pipeline step.

Security3
Quality3
Auditability3
Useful to you4
Useful to community4
Buildable now3
Hermes3

Verdict: watch. Interesting capability but very new; ffmpeg dependency already covered by the existing video-pipeline-tool. Watch for stability before replacing it.

Build #16 video-use: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/browser-use/video-use. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · browser-use GitHub org; published Aug 2026; natural-language edit interface confirmed; ffmpeg dependency stated in README.

#17
watchStandalone tool22 / 35

hermes-agent (NousResearch) repo

NousResearch's Hermes agent: a self-improving agent runtime positioned as an adaptive companion. Tagline is 'the agent that grows with you'.

What it does for you: Interesting benchmark for what a self-improving agent architecture looks like; the name clash with Axion's Hermes is a useful reminder to document Axion's own Hermes scope clearly. No direct integration path.

In practice: NousResearch org; research-grade; adaptive learning angle is novel but vague.

For: Standalone tool. Independent agent runtime from NousResearch; runs standalone, not inside the Axion Hermes bridge.

Security3
Quality3
Auditability3
Useful to you3
Useful to community4
Buildable now3
Hermes3

Verdict: watch. Interesting research direction; no immediate Axion fit. Watch for concrete capability claims backed by benchmarks.

Build #17 hermes-agent (NousResearch): use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/NousResearch/hermes-agent. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · NousResearch GitHub org; adaptive agent architecture described in README; no concrete benchmark results at time of scouting.

#18
newwatchClaude (Axion)22 / 35

scientific-agent-skills skill

Agent skill library targeting 170K+ scientific researchers. Covers literature review, data extraction, hypothesis generation, and experiment planning.

What it does for you: Patterns from research workflows may translate to MCL content research and IFPAS podcast production: literature triage and hypothesis framing are directly applicable to market research steps.

In practice: K-Dense-AI org; large stated audience; skill format compatible with Claude Code.

For: Claude (Axion). Claude Code skill files for scientific research workflows; Claude Code native.

Security4
Quality3
Auditability3
Useful to you2
Useful to community5
Buildable now3
Hermes2

Verdict: watch. Large audience signal but research-domain focus is a stretch for Axion's primary use cases. Some patterns worth borrowing but not a direct build.

Build #18 scientific-agent-skills: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/K-Dense-AI/scientific-agent-skills. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · K-Dense-AI GitHub org; 170K researcher audience stated; skill file format confirmed Claude Code compatible; literature review skill inspected.

#19
watchStandalone tool21 / 35

opencode repo

Open-source coding agent by anomalyco. Provides a self-hosted Claude Code alternative with configurable tool access.

What it does for you: Potential fallback or secondary coding-agent runtime if Claude Code is unavailable; not a priority while Claude Code is the primary.

In practice: anomalyco org; self-hosted angle is appealing for data-sensitive tasks; feature parity with Claude Code is unverified.

For: Standalone tool. Open-source coding agent runtime; runs standalone, not embedded in existing Axion tooling.

Security3
Quality3
Auditability3
Useful to you3
Useful to community3
Buildable now3
Hermes3

Verdict: watch. Interesting self-hosted fallback but feature parity unconfirmed. Claude Code remains primary; revisit if a gap appears.

Build #19 opencode: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/anomalyco/opencode. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · anomalyco GitHub; self-hosted coding agent described; configurable tool access confirmed; feature completeness unverified at scouting time.

#20
newwatchStandalone tool17 / 35

Herdr YC S26 repo

Herdr is an open agent runtime accepted into YC S26. Positions as a standardised layer for deploying and managing production agent fleets.

What it does for you: Early signal on where open agent runtime infrastructure is heading. YC backing means it will mature quickly; worth watching as a potential future Axion deployment layer.

In practice: Pre-product YC announcement; no runnable code yet. Strategic signal only.

For: Standalone tool. Open agent runtime product entering YC; not yet integrable into Axion.

Security3
Quality3
Auditability2
Useful to you2
Useful to community3
Buildable now2
Hermes2

Verdict: watch. Too early to build; YC backing makes it worth tracking quarterly.

Build #20 Herdr YC S26: use the ai-implementation-build-intake skill to build this safely. Source: https://herdr.dev/blog/herdr-is-joining-y-combinator/. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · Herdr.dev blog post; YC S26 confirmed; open agent runtime positioning stated; no code repository linked at scouting time.