Skip to content
View aisona-lab's full-sized avatar
☺️
cooking
☺️
cooking
3 followers · 2 following

Achievements

Achievement: Pull SharkAchievement: Pair ExtraordinaireAchievement: YOLO

Achievements

Achievement: Pull SharkAchievement: Pair ExtraordinaireAchievement: YOLO

Block or report aisona-lab

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
aisona-lab/README.md

aisona-lab

AI engineer building deterministic harnesses around agents.

Thesis → trailer

Role Repo Idea
Thesis / auth agent-action-gate Deterministic allow / deny / human-approval for tool calls. Zero LLM in the auth path.
Trailer lazycoder Rubric code review (APPROVE / REQUEST_CHANGES / BLOCK). Offline prove & replay without a key; live review optional.

Related: skill-guard — static pre-install audit for Agent Skills.

Hiring skim

Thesis → trailer: agent-action-gate is the auth thesis: local-first, deterministic allow / deny / human approval, with zero LLM keys in decide(). lazycoder is the optional trailer: rubric review with deterministic replay and optional live review; neither repo is a hard dependency of the other.

Proof → security: The gate's offline proof is scripts/prove.sh. aag decide --format sarif exports SARIF, and the SARIF example shows upload to GitHub Security → Code scanning. Stage 3a path/glob covers string-only glob / startswith matching with fixtures; residual: no path normalization, symlink resolution, or .. collapsing.

Honesty boundary: Agent Action Gate is source-install for now; no live PyPI release is claimed. lazycoder's Stage 2 corpus LIVE scoring remains pending credits; the claim here is offline proof / replay. Optional adjacent one-liner: skill-guard — static pre-install audit for Agent Skills.

Delivery loop

Short loop from each repo's LEARNINGS.md: SPEC → PLAN → OK → smallest unit → prove with a command → merge → clean tree. No invented precision.

Profile pins

Pinned on the profile (set in GitHub UI — public API has no updateProfilePins for this token):

  1. agent-action-gate
  2. lazycoder
  3. skill-guard (optional)

UI: Profile → Customize your pins → select the repos above.

Popular repositories Loading

  1. lazycoder lazycoder Public

    A code review agent with senior-level judgement. It interrogates every changed block against a fixed rubric, runs the real checks, and returns a defensible verdict APPROVE / REQUEST_CHANGES / BLOCK…

    Python 2

  2. shield-agents shield-agents Public

    Multi-agent AI cybersecurity scanner: 6 specialized agents for vulnerability detection, threat analysis, compliance checking & auto-remediation

    Python 1

  3. prompt-guard prompt-guard Public

    A security linter for LLM prompts: detect prompt injection, jailbreaks, system-prompt leakage, obfuscation and PII exfiltration before they reach your model. CLI + REST API + TypeScript library + w…

    TypeScript 1

  4. OrcaI OrcaI Public

    Your agent proposes. Orca verifies. You decide. A drop-in verification layer for LLM/agent outputs: grounding, schema, PII, rubric checks + retry/repair/escalate and a tamper-evident audit trail.

    Python 1

  5. skill-guard skill-guard Public

    Audit Agent Skills before install. Deterministic checks. Exit codes. No vibes.

    Python 1

  6. agent-action-gate agent-action-gate Public

    Deterministic policy gates for AI agent tool calls: allow, deny, or require human approval.

    Python 1