predictive-mind

Plugin predictive-mind from marketplace jasonhch-plugins in jason-hchsieh/marketplace. Commands read from the repo's manifest — 1 month ago.

Skills (5)

• checkpoint

  Snapshot the current belief state (and optionally git-stash the working tree) under a label, so a risky action can be rolled back if it produces high surprise or breaks acceptance tests.

• plan-efe

  Given a goal and a set of candidate next actions, rank them by Expected Free Energy (pragmatic + epistemic value) using past similar episodes, and pick the winner. Delegates scoring to the policy-selector subagent.

• predict

  Record an explicit predicted observation before a side-effecting tool call. Use whenever you are about to run a command, edit a file, or invoke any tool whose outcome you have not already observed. Upholds the predict-before-act invariant.

• probe

  Run a cheap, reversible real probe to ground a belief instead of hallucinating. Use when dreamer confidence is low or when you catch yourself guessing about repository state.

• reflect

  Surprise-triggered reflection. When a PostToolUse hook reports high prediction error (surprise >= 0.6), invoke /reflect to update beliefs via the belief-reviser and decide whether to continue, replan, reclarify, or ask the user.

Subagents (5)

• belief-reviser

  Perceptual-update role. Given a surprise report (prediction vs actual observation) and the current beliefs, revises the belief entries for the task. Has NO filesystem or shell access — it can only read predictions and write beliefs via the belief-store MCP server.

• cartographer

  Builds the initial task card (z_0) for a coding task by mapping the repo, identifying in-scope files, and enumerating relevant tests. Invoke at task start or after a major scope change. Read-only; does not modify files.

• dreamer

  Counterfactual prediction for a proposed action. Given a task card and a candidate action (diff, command, or tool call), predicts the next observation — which tests will pass/fail, what errors appear, and the blast radius. Read-only; does NOT execute actions.

• hierarchy-arbiter

  Decides whether a string of unresolved prediction errors should stay at the edit-level (keep trying), escalate to re-planning, or escalate to asking the user. Pure judgment — no file, shell, or belief-write access.

• policy-selector

  Scores a set of candidate next actions by Expected Free Energy (EFE) — pragmatic value (goal progress) plus epistemic value (information gain) — using the current belief state and past similar episodes. Returns a ranked list with rationale. Has NO filesystem access; only calls belief-store MCP tools

Hooks (2)

• SessionStart

  bash "${CLAUDE_PLUGIN_ROOT}/scripts/bootstrap.sh"

• PostToolUse

  node "${CLAUDE_PLUGIN_DATA}/belief-store/dist/post-tool-use.js"

Similar plugins