Recover the cycle.

S.E.R.E. is the military acronym for Survival, Evasion, Resistance, Escape — what soldiers train for when the plan stops working. The cycle stalls in known ways; each one has a known move. The discipline is naming the situation early enough that the recovery is cheap.

Context poisoned

Situation. The agent is confused, repeating itself, or drifting from the plan. Tool calls go in circles. Replies start ignoring OPERATION.md and inventing their own version of the op.

Move. Run /clear to wipe the chat context, then re-invoke the phase skill you were in — /warno, /opord, /splash, /chop, or /recon. The skill re-reads OPERATION.md and re-establishes the phase officially; the agent picks up where the last checkpoint left it. Fixing a poisoned context inline almost always costs more than re-booting through a skill.

Change a just-shipped task

Situation. The splash landed but you want to change something — a rename, a missing step, a wrong util, or a new direction entirely.

Move. Branch on size.

• Minor. You’re still in the splash phase from that splash, so just chat the tweak — “rename that helper”, “the AAR missed the migration step, add it”. The agent will make the change and refresh the AAR to match.
• Major. Invoke /opord ... to insert a new TASK and ship the change as its own splash. Re-splashing the same task would mix two stories; a new TASK gives the change its own review window.

Stuck on a call

Situation. A Constraint, a TASK, or a pending change doesn’t sit right. You feel the call is wrong but can’t yet name the fix — or you can’t tell what fix the agent would even agree with. Re-running /warno or /opord blind would just churn.

Move. Drop into /recon for a read-only research-and-discussion phase. Pull the relevant files, talk through tradeoffs with the agent, share the pain you can’t yet name. The agent suggests alternatives; you suggest your own. Nothing lands on disk yet. When you have a concrete prompt, tell the agent /warno ... (for direction changes), /opord ... (for task changes), or combine them — run /warno and /opord to update related Constraints and TASKs — when the idea has solidified enough to touch both layers. The discussion hands off into the right phase skill.

Task conflicts with Constraints

Situation. You ask the agent to reshape a task — split it, merge two, rewrite the approach — and the agent pushes back: the change would contradict a current Constraint. The task can’t move without the WARNO moving first.

Move. Review the pushback. The agent will print a drafted /warno "..." prompt as part of the response — the WARNO revision it thinks is needed. From there:

• If the drafted prompt looks right, tell the agent run /warno and /opord for me — it’ll revise the WARNO, then re-cut the affected tasks under the new Constraints.
• If the draft needs tweaks, copy it, edit it, and submit it yourself. The agent runs against the prompt you actually wanted. /opord follows after the WARNO has settled so OPERATION ORDER re-aligns with the new direction.

Scope ballooned

Situation. The CONOPS is bloated, the Constraints have multiplied past the point you can hold them in your head, or the task list is doing the work of two operations. These are the concrete signs the op should split.

Move. Run /chop ... to draft a split proposal at notes/CHOP.md. Review the proposal (or hand-edit it), then either /chopchop to apply the split and spawn a new OP, or /nochop to discard it. One op, one cohesive ship — bigger isn’t better.

Plan is unsalvageable

Situation. Every checkpoint surfaces more problems than the last. The cycle is digging in instead of moving. You can feel the sunk cost growing.

Move. Drop into /recon and walk through what is worth keeping — research notes, design decisions, partially-shipped work that deserves a second life. Tell the agent to spin up a new OP and post the carved intel as logs on it (MCP tools handle the create + log calls). Then /exfil the dying OP. The plan dies; the lessons live on a fresh start. Sunk cost is the worse cost. Restart cheap.