v0.3.0

Six questions instead of five. Epistemic asks whether what the agent concluded is true; the new plugin asks whether that conclusion reached the reader in a form they can act on. The siblings now hand their open choices to that format, and the skills that used an English phrase list as the trigger name the act first so the protocol blocks can be written in any language.

Plugin	Version
executive-self-monitoring	1.3.2	Am I doing what the plan asks?
epistemic-self-monitoring	0.1.3	Is what I concluded actually true?
persistence-self-monitoring	0.1.3	Is it still worth insisting on this?
termination-self-monitoring	0.1.3	Is this stop justified by something checkable?
coverage-self-monitoring	0.1.3	Did I deliver every part, including the hard one?
handoff-self-monitoring	0.1.1	new — Can the reader act on what I wrote?

New: handoff-self-monitoring

The agent knows the state, the problem and the open decision, and writes its close in the register of its own trace — paths, identifiers, what it ran. The reader has the message, not the trace, and cannot tell what to decide or what to do next (curse of knowledge, illusion of transparency, writer-based prose). The [HANDOFF] block is modelled on SBAR / I-PASS: Status (done | needs-decision | blocked), Situation in the reader's terms, Options with a Default when there is a fork, Blocked-by when blocked, Next as one action or nothing.

Three layers: load on the first prompt and every 10th; a pre-close nudge on the first closing-shaped tool call of the turn (a green test / build run, a commit or push, a PR opened), so the format arrives when the final message is about to be written; a Stop scan for a decision named but not handed off (an offer, a fork, a question in the last 6 lines, a returned coverage part) and for the block's rules. Non-blocking by default (a one-line retrospective on the next prompt); HANDMON_STRICT=1 turns the Stop into a gate that fires once. Claude Code: UserPromptSubmit, PostToolUse, Stop. Cursor: sessionStart, postToolUse, afterAgentResponse, stop.

Also

• coverage's returned part, termination's owner-choice, persistence's Report and epistemic's closure now point at the [HANDOFF] block, so a choice reaches the owner formulated (options, consequences, a default) rather than named.
• handoff 0.1.1, coverage 0.1.3, termination 0.1.3 — skill and injected reminders name the act first (an offer, a hole in the delivery, a stop on a feeling). English phrases stay as examples and as the hook lexicon backstop. Block field names stay in English; values and the rest of the message follow the language of the turn. scripts/test.js locks that contract: English lexicon hits still fire, Spanish semantic equivalents do not, and an English-keyed block with Spanish values still passes.
• SECURITY: three opt-in gates (handoff's HANDMON_STRICT joins epistemic's and termination's). Issue templates list all six plugins. Social preview updated to the six questions.

Install / update

claude plugin marketplace add 3dgiordano/agent-plugins
claude plugin install handoff-self-monitoring@3dgiordano-agent-plugins
claude plugin update coverage-self-monitoring@3dgiordano-agent-plugins
claude plugin update termination-self-monitoring@3dgiordano-agent-plugins
claude plugin update epistemic-self-monitoring@3dgiordano-agent-plugins
claude plugin update persistence-self-monitoring@3dgiordano-agent-plugins

Cursor: add the repo as a marketplace and install (or update) from Customize → Plugins. Restart Cursor after installing.

Full changelog: CHANGELOG.md · v0.2.0...v0.3.0