Fulcrum 9 of 18
While it works
During long stretches of work between messages, and whenever several agents share one machine.
What goes wrong here
The agent narrows. It grinds on one hypothesis that isn't working, loses the protocol it started with, drifts into dense technical narration, or quietly moves to a different sub-problem without saying so. When agents run in parallel, one can overwrite another's merge or take over a port the person is using.
How it compounds if nothing catches it
Narrowed effort feels productive while it moves away from the goal, and the longer it runs, the more work there is to throw away. Silent scope shifts mean the person's picture of what the agent is doing stops matching reality. Collisions between agents destroy work outright.
What the harness does here
expand-perspective is a frame-exit ritual, with 100 worked examples. When something is taking longer than expected, the agent stops at intervals to ask whether the frame it is working in is the limit. loop keeps a task going in rounds until it is proven done or blocked by a named outside dependency, so the person never has to say 'keep going'. calibrate re-reads the operating protocol when the agent has drifted. on-scope-change updates a visible breadcrumb (deepest task < parent < root) every time the focus shifts. smc watches for signs of drift, such as the person's frustration rising above their own baseline, and injects corrections that have worked before. Small guard hooks block a second merge into a shared branch while one is running (pre-tool-merge-lock.sh), and stop agents from taking a port the person keeps for their own use (pre-tool-founder-port-guard.sh).
The pieces that act here
- /expand-perspective
Frame-exit ritual for when you're stuck. If you have been unable to effectively solve a challenge within the amount of time you expected it would take to solve it, start asking every 5 minutes whether you are operating from within a perspective that is in fact limiting you. Teaches the perspective-shift move — popping your head out of the current frame to check whether there is more signal in a wider view — through 100 keyhole→wide-view examples across every domain. Use when debugging drags, when hypotheses keep failing, when effort exceeds expectation, or any time progress feels productive but the problem won't resolve.
- /loop
Persistent completion loop for any objective that must be finished, not attempted. Arm a tracked goal, work in rounds, run a frame-exit check whenever stuck, and stop only when done is proven with runnable evidence — or when the same concrete external blocker has persisted across rounds with the missing dependency named exactly. Use when the human says "loop", "keep going until done", "get unstuck if stuck", "don't stop until it works", "until done and proven", or when YOU notice a task is long-running, multi-scope, or you are about to hand back work that is incomplete or unverified. This skill exists so the person you're working with never has to remind you to keep going — invoke it yourself.
- /calibrate
Re-ground to the harness's core operating contract. Invoke when an agent is drifting from its core operating contract — communicating too technically/densely, skipping coherence checks, not reflecting intent before acting, or losing track of the alignment-first principle. Re-injects the contract below directly into context so the agent remembers what it's actually supposed to be doing. Especially important at end of substantive turns where agents lose protocol awareness amid context bloat.
- /on-scope-change
Use whenever the active scope shifts during a session — entering a sub-task, dropping into a bug, climbing back to a parent goal, or pivoting. Emits a visible nested breadcrumb of the current focus (deepest < parent < root) so the person you're working with gets instant recall of exactly where the agent is in the work's nesting, and fires the chapter-mark (if that tool is available) so it lands in the transcript/TOC.
- /smc
Alignment-drift injection — detect when a session is drifting from the person's stated intent (rising frustration, repeated corrections, a high-risk task with no completion check) and surface a targeted past correction, rather than repeating the same mistake pattern silently. Use /smc to check status, verify telemetry, or troubleshoot.
- /smc-frustration-detector
A reference pattern for detecting frustration spikes in a person's messages and logging a telemetry event when they happen. Wire it into your own project's UserPromptSubmit hook if you want it running on every message — it is not wired in automatically by installing this skill.
- pre-tool-merge-lock.sh
A hook: a script Claude Code runs on its own at a set moment.
- pre-tool-founder-port-guard.sh
A hook: a script Claude Code runs on its own at a set moment.