context-grapple-gun

Context Grapple Gun

Context Grapple Gun is a portable governance lifecycle for Claude Code and a developer-facing entry point into Prompted LLC's Ubiquity substrate.

Prompted LLC builds Ubiquity so AI systems can know when to act, when to ask, and how to turn human judgment into safer autonomy over time.

CGG applies that thesis locally: it captures lessons from real work, routes them through human review, promotes them through scoped gates, and hydrates approved guidance back into future sessions without confusing rendered context with constitutional source of truth.

CGG is complete without Ubiquity. When flat-file governance is no longer enough, Ubiquity extends the same primitives into deeper substrate infrastructure: semantic recall, graph topology, conformation-aware retrieval, expression gating, and compiled execution-boundary enforcement.

Context Grapple Gun, also published as the open-source context-grapple-gun repository (prompted365/context-grapple-gun), is the developer-facing implementation of this portable lifecycle.

Three commands. Auditable files. No service required.

---

You only need four things to start:

• Three commands: /cadence, /review, /siren
• Run /cadence to end every session cleanly; run /review every few sessions to approve lessons
• Scope ladder: Site > Domain > Estate > Federation > Global -- you decide what promotes
• Install: npx context-grapple-gun install -- or see INSTALL.md for details

---

Read this first (by intent)

You want to...	Start here
Install now	npx context-grapple-gun install -- zero-thought entry, handles everything
Use it now	START-HERE.md -- the three commands, a normal day, what to expect
Understand the pipeline	DEV-README.md -- session flow, hook lifecycle, extraction pipeline
Evaluate the architecture	ARCHITECTURE.md -- signal manifold, acoustic model, governance layers
Audit recent changes	docs/COMMIT-HISTORY-CHEATSHEET.md
Learn through story	academy/README.md -- five chapters, real simulations, one very persistent goat
All install options	INSTALL.md -- npm, plugin, bootstrap, manual, academy

---

90-second mental model

The problem: AI agents discover truths during work -- bug patterns, API quirks, coordination techniques that work. When the session ends, that knowledge vanishes. Next session: same agent, same repo, zero memory of what it learned.

The CGG answer: Lessons get captured as they happen, reviewed between sessions, and promoted to broader scopes with human approval at every gate. The project's operating rules grow from real work, not from someone writing documentation.

Three commands run the lifecycle:

Command	What it does
/cadence	End of session. Saves lessons, emits a tic (sequenced timestamp), writes a handoff for the next session.
/review	Review proposed lessons. Approve, reject, or modify before promotion.
/siren	Check on recurring friction. See what signals are building, what warrants have minted.

Why this compounds instead of just accumulating:

Mechanism	What it does
Scoped gates	Every promotion requires explicit approval. The agent proposes; you decide what scope it reaches.
Abstraction ladder	Scope hierarchy: site > domain > estate > federation > global. Lessons climb through review, not drift.
Hydration boundary	Rendered working context is never confused with constitutional source of truth. Lessons hydrate back into sessions from their authoritative files.
Cognitive pull requests	Proposed lessons are captured as CogPRs — structured, reviewable units that strip contextual noise before promotion.
Signal manifold	Recurring friction accrues volume, crosses thresholds, mints warrants. You see what keeps coming up.

One scale boundary:

CGG is the governance lifecycle. It uses flat files, git-tracked, auditable by default. When flat files aren't enough -- when you need semantic recall, graph topology, or conformation-aware retrieval -- the substrate layer (Ubiquity) picks up where CGG leaves off. Same governance primitives, deeper infrastructure. CGG is complete without Ubiquity. Ubiquity composes on top when scale demands it.

---

What CGG is not