More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Agent Action Log: tamper-evident audit trail for AI agents

    1

    Your agent logs every step into a signed, hash-chained record, then answers 'what did my agent do yesterday, and was the log changed?' with proof.

    Free

    2 usesSecurity scanned
    Agent Action Log: tamper-evident audit trail for AI agents

    Agent Action Log: tamper-evident audit trail for AI agents

    Example session with this skill installed

    Check release v1.4.4 against the staging health endpoint. Keep a log I can show the client, and tomorrow I will ask what you did.

    • Read your context and instructions
    • Compiled the agent action log

    Summary

    Release v1.4.4 is blocked: staging health returned 503. The run is logged and sealed: 3 signed steps.

    What the agent logged (agentlog MCP server)

    StepCallContent
    0log_action action run.startnote: "Check release v1.4.4"
    1log_action action tool.call, tool http_getinput {"url": "https://staging.example/health"}, output {"status": 503}
    2log_action action run.endoutput {"result": "blocked: health check 503"}

    Only the hashes of input and output enter the chain; the raw data stays on your server as private evidence.

    Next day: "What did release-checker do yesterday, and was the log changed?"

    The agent calls audit_day and reads the answer as is:

    Yesterday, release-checker ran 1 time, 3 logged steps in total. The task was: Check release v1.4.4. Every log is intact: signed by the agent, linked, and sealed.

    If someone rewrites the 503 to 200 in the database

    The same question now returns

    Yesterday, release-checker ran 1 time, 3 logged steps in total. The task was: Check release v1.4.4. Warning: one log was tampered with. In the run at 5:08 PM, "Check release v1.4.4", step 1 cannot be trusted: its content was edited after the agent signed it.

    Next steps

    • get_run returns an export anyone can re-verify offline.
    • If the client and the operator disagree about the 503, diff_versions replays each side's copy against what the agent signed and says who is right.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    What it does

    AI agents now run tools that matter: they approve releases, issue refunds, send messages, change records. What an agent did is usually kept in a log on the operator's server, and after an incident that log is the only evidence, held by the party with the most reason to change it.

    This skill makes your agent keep that evidence properly. It drives the free, open-source agentlog MCP server (MIT): every step of a run (tool call, model call, error, end of run) is hashed, signed with the agent's key and linked to the step before it. If anyone edits, deletes, reorders or re-signs a step afterwards, the verifier says which step and how.

    What the skill teaches your agent

    • When to log: at the start of a multi-step task, after every tool or model call, on errors, and at the end (the end seals the run, so a cut-off run shows as open, never as complete).
    • When to read: "what did my agent do yesterday?", "was the log tampered with?", "is run X intact?", "who is right, the client or the operator?".
    • What it may claim: never "intact" unless a verify tool said so in this conversation; "open" is not "intact"; it proves a change, not who made it.
    • Privacy: no secrets or personal data in the public labels (note, agent_id, run_id). Raw inputs and outputs stay on your server; only hashes enter the chain.

    What is in the package

    • SKILL.md: triggers, the logging routine, the audit routine, verdict wording, boundaries, one worked example.
    • references/setup.md: connect the server in about 3 minutes (Node.js 22.18+, one clone, one start command).
    • references/tool-reference.md: the six tools (log_action, audit_day, verify_run, list_runs, get_run, diff_versions) and their fields.
    • references/examples.md: clean, open, tampered and disputed runs with real outputs.

    Honest limits

    • It needs the agentlog MCP server running (free, MIT, local SQLite). The skill alone does not log anything.
    • Someone who holds both the agent's key and the log file can rewrite a whole run consistently. The server can also write to the public Arkiv network (read-only entries) to close that gap; this is optional and uses a test key.
    • A log cannot prove that an agent told the truth at the moment it wrote.
    • Checked end to end with the official MCP TypeScript SDK client against a live server (log, verify open and sealed run, list, daily audit, export, dispute replay). Not yet tried inside every individual agent listed in the compatibility strip.

    Source, tests (66 passing) and demo video: https://github.com/ttimesai-star/agentlog-mcp

    Needs the agentlog MCP server running; the skill alone logs nothing.
    Anyone holding both the agent's key and the log file can rewrite a whole run consistently (optional Arkiv mode closes this).
    Proves that a step changed, not who changed it, and not that the agent told the truth when it wrote.
    Verified with the official MCP TypeScript SDK client; not yet tried in every listed agent.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    2 uses

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    • Free with an Agensi account

    Requires

    Agentlog (Free Server, MIT) MCP
    Node.js 22.18 or newer
    Agentlog MCP
    Node.js
    Git

    Permissions

    No special permissions declared or detected

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    No. It tells your agent how to use the agentlog MCP server. Connect that server first (references/setup.md, about 3 minutes).

    No, by default. State is a local SQLite file and one key per agent. Writing to the public Arkiv network is optional and only hashes and signatures go there.

    No. It proves that a step was edited, deleted, reordered or re-signed, and which step. It does not identify the author of the change.

    New in LLM Ops