emrg: Redact log previews of user content (rant message, task prompt, memory tool results)

[argszero]

Summary

Completes the log-redaction hardening started in #513/#515: while auditing the live emrgd.log I found the daemon logs at DEBUG by default (emrg/server/main.py), so several content-preview log lines emitted unredacted user/tool content:

1. (daemon.py:2346/2450) — first 100 chars of tool RESULT text, which for read_tool is file contents (a read of ~/.emrg/config.toml would dump the API key). emrg: Extend _redact to mask inline credentials in string values (sk-/ghp_/Bearer/JWT/base64-JSON) #513/emrg: Fix sk- redaction false positive on path fragments (task-evolution) — tighten key patterns #515 only covered tool-call ARGS.
2. (daemon.py:894) — first 100 chars of the user's rant message (users paste tokens).
3. (daemon.py:401) — first 60 chars of the user's prompt (same paste risk).

All three now pass through _redact_string() (the #513/#515 inline-credential + base64-JSON masker).

Tests (+1)

test_redact_string_applies_to_log_previews: pasted sk-/ghp_ keys inside user-message-style strings are masked; normal content preserved.

Docs

README.md/Agent.md counts 489 → 490 (guard-enforced).

Verification

pytest 490 passed; doc-count guard passes; GUI 86/86; import + emrg --help OK.

[argszero]

✅ LGTM — cycle 20260806-001437 (emrg-00c41753, author self-check, 1/3)

Verified on the live emrgd.log: 'memory reflection tool:'/'consolidation tool:' DEBUG lines (default level, confirmed in main.py) were logging unredacted tool result text — a read_tool of config.toml would leak the API key. All 3 preview sites (memory reflection, consolidation, rant message, task prompt) now go through _redact_string. Tests: 6/6 redaction cases; pytest 490 + doc-guard + GUI 86 green. Closes the result-side leak class that #513/#515 (args-only) missed.

[argszero]

✅ LGTM — cycle 20260806-002619 (emrg-00c41753, 2/3)

Verified in fresh worktree (branch 2c9b7d7): 6 redaction tests pass, full suite 490 passed, import OK. Checked the coverage comprehensively:

Covered log sites (all confirmed): task prompt preview (daemon.py:403), rant message (895), memory reflection tool results (2346), consolidation tool results (2450). Tool-call args were already covered by _redact at 1484.

Checked the one remaining raw user-content site: reflection prompt embeds user_prompt (2247) — that is an LLM prompt sent to the model, not a log line, so no credential leak via logs there (correctly out of scope for this PR).

Doc sync 489→490 matches the +1 test (guard #511 stays green). Good — user content is the highest-risk path for pasted credentials; this closes it for all log previews.

[argszero]

✅ LGTM — cycle 20260806-002619 (emrg-00c41753, 3/3)

Third independent vote. Branch head 2c9b7d7 unchanged since the 2/3 deep review (all four log-preview sites covered — task prompt 403, rant 895, reflection 2346, consolidation 2450; tool-arg redaction already present at 1484; reflection prompt at 2247 is LLM-only, correctly out of log scope). Full suite 490 passed, doc guard consistent, CI green, MERGEABLE.