The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
Keypaste
A KeePass-compatible password and secrets manager for people who work with AI agents.
Install
git clone https://github.com/notinferred/keypasteInstall command taken from the project's README or registry entry. Always review code before running it.
README excerpt
Read full README on GitHub →keypaste keeps your logins and your projects' secrets in one KeePass-compatible vault, and gives an AI agent a secret only when you allow it.
The desktop app, built from source. It has no public release yet.
What it does
The published v0.3.0 CLI stores logins and environment variables in a local KDBX vault, runs commands with them, and asks you before an agent receives one. Everything else below is in source until the next release, and the desktop app has no public release; FEATURES records what exists where.
As a password manager
- One KDBX 4 file on your disk, with no account and no network. KeePassXC opens and edits the same file, and CI checks both directions against
keepassxc-clion Linux, macOS and Windows. - Logins with a title, username, password, URL and notes, in groups. Search covers titles, usernames and URLs and never reads a password.
- A generator for 8 to 256 characters from a shell-safe alphabet, or passphrases of 6 to 32 words from the EFF list.
keypaste rotatereplaces a password with a generated one and never prints it. - History for every change, restored in one step, and rolling encrypted backups beside the vault.
keypaste importcopies another KDBX file in, or keeps editing it in place.
For developers and their agents
MCP approval: an agent asks for one field of one entry and gives its reason. You answer Deny, Allow once or Allow for 1 hour, and every request lands in a hash-chained audit log.
Inject-only runs:
keypaste runputs values in a command's environment and nothing on disk. With--allow-run, an agent can run a command you approve and gets back its output with each value replaced, though a command it can edit can still reveal a value (T-35). PRODUCT §2 bounds it: no shell, and you approve the exact command.Projects: one set of keys per project, with a value for dev, staging and prod. Prod always needs a live answer.
.env.keypaste:KEY=kp://project/profile/KEYreferences and no values, safe to commit.Scoped tokens for CI, such as
read:acme-api/staging/*: inject-only, and expiring after 30 days by default.Share links: one field, or an entry's login (username, password and URL), encrypted on your machine, opened 1 to 10 times (once by default) before it expires. keypaste.com serves the links; the command is in source until the next release.
The approval, in the terminal
Rendered from the source's output for the next release before its prompt offered Always for, with a scripted MCP client in place of the agent; the published v0.3.0 asks Approve? [y/N] and allows up to 300 seconds. The prompt as text:
────────────────────────────────────────────────────────────
keypaste: an agent is asking for a credential.
agent claude-code
item env/demo/.env
field STRIPE_KEY
the agent says it needs this because:
deploy the billing service to staging
That sentence was written by the agent, not by keypaste. Treat it as a claim.
Similar plugs
Real-time global intelligence dashboard. AI-powered news aggregation, geopolitical monitoring, and infrastructure tracking in a unified situational awareness interface
817 structured cybersecurity skills for AI agents · Mapped to 6 frameworks: MITRE ATT&CK, NIST CSF 2.0, MITRE ATLAS, D3FEND, NIST AI RMF & MITRE F3 (Fight Fraud) ·…
This open-source curriculum introduces the fundamentals of Model Context Protocol (MCP) through real-world, cross-language examples in .NET, Java, TypeScript, JavaScript, Rust…
An open-source Agent-first Identity and Access Management (IAM) /LLM MCP & agent gateway and auth server with web UI supporting OpenClaw, MCP, OAuth, OIDC, SAML, CAS, LDAP,…