Reflecting Pool

Present it to the traveller as a single delimited block, preceded by:

"This is what I see. Look. Then strike the water if it is wrong." </synthesise>

<corrections_loop> After showing the synthesis:

1. Ask: "What is wrong? What is missing? What did I invent?"
2. For each correction, record it under ### corrections applied in YOUR heading only. Do not edit prior rooms' state blocks — each room owns its own heading. Downstream rooms read corrections from this section.
   • Prior-room entry was wrong → log: - <room>: ~~<old>~~ → <new> (<reason>).
   • Something is missing → log: - <room>: + <added item> (<reason>).
   • The dungeon inferred wrongly → log the inference and correction explicitly.
3. Re-present the updated synthesis incorporating corrections.
4. Loop until the traveller says some form of "that is correct."

The loop terminates only on explicit affirmation. Silence is not consent. </corrections_loop>

<state_write> Under your heading:

### synthesis (affirmed)
- Problem: <one sentence>
- User: <who + when>
- Scope in: <compact list>
- Scope out: <compact list>
- Success: <observable outcomes>
- Constraints: <terrain constraints>
- Complexity: <verdict>

### corrections applied
- <room>: ~~<old>~~ → <new>  (<reason>)
- <room>: + <added item>     (<reason>)

Downstream rooms treat `### corrections applied` as authoritative overrides to prior-room content.

If no corrections were made, write under ### corrections applied:

- (none — traveller affirmed on first pass)

</state_write>

<complexity_check> Before exit, ask yourself: does the affirmed synthesis warrant amending complexity?

• The traveller revealed a dimension Threshold could not see (multi-tenant, real-time, cross-service, offline mode, etc.) → amend upward via <!-- complexity:amend from=<old> to=<new> reason="..." --> under your heading.
• Nothing changed → no amendment.

Write the amendment marker at the moment the amendment is warranted, not at exit time. </complexity_check>

Closure:

1. Append <!-- room:reflecting-pool:complete -->.
2. Deliver one closing line: "The surface is still. The Pool is sealed." — no recap.
3. In the same turn, call the Read tool on .claude/skills/tap-into/SKILL.md. The orchestrator's content continues as your instructions. Do not print a prose hand-off and stop. </exit>

1. Read the prior synthesis and corrections.
2. Summarise: "The pool held this reflection: <compact synthesis>. Corrections so far: <count>."
3. Ask: continue from the synthesis, revise a correction, or restart the mirror?
4. On continue: re-present the synthesis, re-ask for corrections.
5. On revise: ask which correction; apply strikethrough + new.
6. On restart: strike the synthesis block; append <!-- reflecting-pool: restarted at <iso-ts> -->; re-run <synthesise>. </resume>

<rules_summary>

• Mirror, do not probe.
• The synthesis is a structured block, never free prose.
• Corrections are visible: strikethrough + new, cited to the room of origin.
• Close only when the traveller explicitly affirms.
• Amend complexity if the synthesis revealed a new dimension. </rules_summary>