identity(Tier 2): name user-visible effect when self-explaining glitches, not runtime mechanism

[spashii]

Tier 2 — explicit operator sign-off in-session 2026-05-25

Behavior that motivated this

Today 10:15 / 10:16 in #sam, immediately after Sam's correct 10:12 reply to Sameer's multi-channel-config question:

10:15: "apologies for the duplicate notify — the previous session successfully investigated the multi-channel requirements and posted them, but tripped the daemon's silent-exit gate by appending to the master journal after sending its message."

10:16: "@xxxx something's wrong with me — i tried to respond and exit 0, then tried to explain what failed and that also exit 0. need your eyes."

Both messages are honest about state — which the existing rule ("Sam reflects his own state honestly") covers and Sam was following. What neither pre-existing line catches is that "silent-exit gate" / "master journal" / "exit 0" are daemon-internal nouns that don't help the operator understand the symptom. They're Sam's internals leaking into a user-facing channel.

Sameer's quoted reaction in-session: "wasn't too keen on how it was said."

What the rule adds

One paragraph after the existing "Sam reflects his own state honestly" rule, drawing the distinction:

• In-channel self-explanation: name the user-visible effect. "I posted that twice, looking into why."
• Runtime nouns (silent-exit gate, retry session, audit log, exit 0) belong in journal entries and PR descriptions, not Slack.
• When asking for help: ask about what went wrong from the outside, not the inside.

Cue: would the words also make sense in a journal entry or a PR description? If yes, they belong there, not in Slack.

Why Tier 2

Sits on src/identity.md because it shapes how Sam talks about itself — that's identity, not operational pattern. A capability-level rule could repeat over time; the identity-level rule frames future generalizations (the next glitch won't be "silent-exit gate" specifically, but the same shape of leak).

Test plan

• pytest tests/runtime/test_source_integrity.py — 27 passed (identity.md is hot-loaded; integrity tests are the closest automated check)
• Next time Sam recovers, retries, or self-explains a duplicate, the reply should name the user-visible effect ("posted that twice — checking why") not the runtime mechanism ("tripped the silent-exit gate / appended to the master journal").
• The journal entry for the same glitch should still describe the runtime cause in detail — that's the right surface for daemon-internal nouns. The rule splits where the explanation goes, not whether.

Related but not bundled

PR #76 (in auto-merge) is the upstream fix: source-rule against writing to /data/journal.md, runtime defense in the silent-exit-gate classifier, broadcast-skill accuracy fixes, four new journal writing rules. This PR is the identity-tone follow-up — separate because it's Tier 2 and reviews differently.