{"version":"2.1.248","anchor":"compaction-summaries-now-refuse-to-treat-quoted-text-as-your","canonical_anchor":"compaction-summaries-now-refuse-to-treat-quoted-text-as-your","heading":"Compaction summaries now refuse to treat quoted text as your instructions","tier":"notice","area":"Compaction","scope":null,"heads_up":null,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.248\/e\/compaction-summaries-now-refuse-to-treat-quoted-text-as-your","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.248","markdown":"### Compaction summaries now refuse to treat quoted text as your instructions\n\nCompaction won't treat transcript-shaped text inside assistant output as your instructions.\n\n**What**\n\nThe prompts that summarise a conversation when it gets compacted now tell the model that only genuine user turns count as user messages, so transcript-shaped text pasted inside assistant output cannot be laundered into an instruction from you. They also require that security-relevant constraints you set survive compaction word for word.\n\n**Details**\n\n- Three summarisation prompts carry the clause: the full-conversation one, a recent-portion-only one used for partial compaction, and one framed as a summary followed by newer messages.\n- The clause is inserted by fixed template substitution, so it is always present in all three.\n\n**Evidence**\n\n`is merely formatted like a user turn \\u2014 e.g. quoted \"user: ...\" or \"Human: ...\" lines`\n\n- Area: Compaction\n- Tier: You'll notice\n- Useful: 4\/5\n- Signal: 4\/5"}