Commit Graph
761 Commits
Author SHA1 Message Date
ciregenzandClaude Opus 5 fb92d785b4 [eric] agents: a human's Stop outranks every recovery path, checked before any of them (ENG-402, ENG-369)
The liveness door resurrected stopped sessions; a stopped browser agent looks exactly like a dead sidecar.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018foyDoK19jjbYdudfzQVkZ
2026-08-24 22:59:44 -07:00
ciregenzandClaude Opus 5 10084b1b28 [eric] agents: split the delegation backstop and the quit telemetry into their own files
Both host files crossed the 300-line ceiling; the seams were already there (quick-tool wedge vs delegation backstop, ladder logic vs its telemetry).

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018foyDoK19jjbYdudfzQVkZ
2026-08-24 22:07:31 -07:00
ciregenzandClaude Opus 5 b2ae28bd8c [eric] agents: record the one nudge outcome the user actually sees (ENG-399)
1,695 nudges were counted on one install and zero outcomes, so "how often does the recovery fail" was unanswerable.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018foyDoK19jjbYdudfzQVkZ
2026-08-24 22:02:03 -07:00
ciregenzandClaude Opus 5 8761320213 [eric] agents: a frozen sidecar can no longer hang a delegation forever (ENG-402)
Exempt from the 25s wedge deadline was silently also exempt from the 300s ceiling; the discriminator is liveness, not time. Reported by Haik Decie.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018foyDoK19jjbYdudfzQVkZ
2026-08-24 21:27:13 -07:00
ciregenzandClaude Opus 5 dc014c5377 [eric] skills: an export never ships a venv, and a licence id is not an API key (ENG-401)
Both halves diagnosed by Haik Decie.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018foyDoK19jjbYdudfzQVkZ
2026-08-24 11:38:26 -07:00
ciregenzandClaude Opus 5 2e8d431118 [eric] agents: a run's result is not a replay, so the final answer survives the trail (ENG-396)
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018foyDoK19jjbYdudfzQVkZ
2026-08-24 11:16:06 -07:00
ciregenzandClaude Opus 5 264f2ef6de [eric] agents: the nudge ladder's last rung is bounded by construction, and can never widen a ratchet (ENG-399)
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018foyDoK19jjbYdudfzQVkZ
2026-08-24 10:54:05 -07:00
ciregenzandClaude Opus 5 5d3eb5cc52 [eric] agents: no renderer bound for a model replays another agent's turns (ENG-396)
Four paths still emitted the USER:/ASSISTANT: shape ENG-358 removed from the recap, because they were tool results and aux calls rather than the recap.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018foyDoK19jjbYdudfzQVkZ
2026-08-24 10:30:36 -07:00
ciregenzandClaude Opus 5 596c0479ca [eric] agents: a ceiling on harness-started turns, so a runaway loop cannot spend an account (ENG-398)
Measured across 79 installs: median 1/min, 76 of 79 under 10/min, and one install at 186/min that produced 44 of 45 policy blocks.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018foyDoK19jjbYdudfzQVkZ
2026-08-24 10:22:54 -07:00
ciregenzandClaude Opus 5 482e94ecb2 [eric] tests: one shared logger capture, and the shaper's names follow the p_ convention
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018foyDoK19jjbYdudfzQVkZ
2026-08-24 01:06:56 -07:00
ciregenzandClaude Opus 5 f8d67fdfd1 [eric] release: 1.7.9 ships with a story, and two log assertions stop reading the root logger
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018foyDoK19jjbYdudfzQVkZ
2026-08-24 01:04:59 -07:00
ciregenzandClaude Opus 5 3e0e1f67c6 [eric] agents: a malformed request is never transient, whatever reset hint rides along (ENG-395)
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018foyDoK19jjbYdudfzQVkZ
2026-08-24 00:48:51 -07:00
ciregenzandClaude Opus 5 626d9ea4dd [eric] agents: bound what one tool result costs the model, without ever deleting the answer (ENG-385)
The 50KB cap shaped our own transcript copy, never the model's context, so it could not have cut a token at any threshold.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018foyDoK19jjbYdudfzQVkZ
2026-08-24 00:45:37 -07:00
ciregenzandClaude Opus 5 cf5093d24f [eric] agents: a process that will inject failures says so, and names a typo that armed nothing
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018foyDoK19jjbYdudfzQVkZ
2026-08-23 21:12:06 -07:00
ciregenzandClaude Opus 5 dc807e09ef [eric] agents: a drilled fault must be the one the real classifier catches, and fire once (ENG-382)
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018foyDoK19jjbYdudfzQVkZ
2026-08-23 21:09:23 -07:00
ciregenz b047e31abd [eric] agents: a refusal guard may not eat a real answer; wording alone never destroys content (ENG-387) 2026-08-23 15:27:36 -07:00
ciregenz 8ed2330c76 [eric] agents: the two guards that could fail silently now declare themselves (ENG-388, ENG-386) 2026-08-23 13:19:26 -07:00
ciregenz d4bfc122f0 [eric] agents: OSW_FAULT fires real provider failures on demand so the guards get drilled, not hoped for 2026-08-23 11:31:58 -07:00
ciregenz 7e4bb7a43b [eric] agents: OSW_COMPACT_CEILING_TOKENS makes the mid-turn breaker drillable without a 180K turn 2026-08-22 21:55:05 -07:00
ciregenz df31e8d5d2 [eric] tests: the suite can never resume real chats with live credentials (ENG-388) 2026-08-21 11:51:14 -07:00
ciregenz 6cc705c0ee [eric] models: only the catalog retires a pinned model, so a router bounce cannot move a chat vendors (ENG-386) 2026-08-21 11:51:14 -07:00
ciregenz 2c8ae9a3b8 [eric] agents: extraction-shaped handoffs and relayed refusals never reach the subscription lane (ENG-387) 2026-08-21 11:51:14 -07:00
ciregenz 2583c80f6c [eric] agents: a session about to continue reports running, and releases it if the continuation dies (ENG-390) 2026-08-21 11:51:14 -07:00
ciregenzandClaude Opus 5 771a7243b0 [eric] agents: a silent quit rebuilds only when the rebuild would buy something (ENG-354)
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018foyDoK19jjbYdudfzQVkZ
2026-08-21 10:29:56 -07:00
ciregenzandClaude Opus 5 da18c0a5ef [eric] agents: the Resume chip is a human, so a stopped chat honours it (ENG-384)
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018foyDoK19jjbYdudfzQVkZ
2026-08-21 08:07:55 -07:00
ciregenzandClaude Fable 5 de70e9fe79 [eric] agents: a subscription-lane policy block fails over to the user's own API key (ENG-383)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01En8dRGsJPLrJCQBEkTH4Mp
2026-08-21 00:20:05 -07:00
ciregenzandClaude Fable 5 42e4f9116b [eric] agents: transport heals respawn the CLI on the same transcript instead of rebuilding it (ENG-382)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01En8dRGsJPLrJCQBEkTH4Mp
2026-08-21 00:20:04 -07:00
ciregenzandClaude Fable 5 a4be8c469f [eric] agents: the recap carries no model text, nudges are short and unlabeled (ENG-358)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01En8dRGsJPLrJCQBEkTH4Mp
2026-08-21 00:20:04 -07:00
ciregenzandClaude Fable 5 295a7b52e2 [eric] agents: a policy block ratchets the recap down instead of bricking the chat (ENG-358)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01En8dRGsJPLrJCQBEkTH4Mp
2026-08-20 23:44:44 -07:00
ciregenzandClaude Fable 5 e8dc7c5fdd [eric] import: only markdown becomes a bare skill, and skills always confirm (ENG-376)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01En8dRGsJPLrJCQBEkTH4Mp
2026-08-20 22:22:49 -07:00
ciregenzandClaude Fable 5 5e0e89abe1 [eric] agents: a fresh heartbeat extends the unwedger to a 300s hard ceiling (ENG-368)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01En8dRGsJPLrJCQBEkTH4Mp
2026-08-20 22:22:49 -07:00
ciregenzandClaude Fable 5 e90abb7bfc [eric] system: watchdog deaths blame no workflow, card tails not owed, 20s probe (ENG-366)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01En8dRGsJPLrJCQBEkTH4Mp
2026-08-20 22:22:49 -07:00
ciregenzandClaude Fable 5 1f1ce09d7a [eric] agents: 401 needs auth context; answers can be blocks; quits forgiven (ENG-364/365)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01En8dRGsJPLrJCQBEkTH4Mp
2026-08-20 22:22:49 -07:00
ciregenz 3235324122 [eric] agents: a user's Stop or Close is final; no watchdog resend, disk reload, read, or late-booting child revives the card 2026-08-20 19:39:33 -07:00
ciregenzandClaude Opus 5 6edce86d68 [eric] browser: children survive turn 1 on every lane, never report "done" for nothing, tables render first try
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014wtspwSFzZmjCx9UNPAorQ
2026-08-20 18:16:51 -07:00
ciregenz b14791074f Revert "[eric] agents: Stop disarms every queued injection, so a stopped agent can never resurrect"
This reverts commit 332ae7d183.
2026-08-20 16:13:50 -07:00
ciregenz 10bca6ba17 Revert "[eric] browser: a card the renderer parked is slow, not dead; parallel agents no longer ghost"
This reverts commit 6ed6206c7b.
2026-08-20 16:13:50 -07:00
ciregenzandClaude Opus 5 6ed6206c7b [eric] browser: a card the renderer parked is slow, not dead; parallel agents no longer ghost
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014wtspwSFzZmjCx9UNPAorQ
2026-08-20 15:50:55 -07:00
ciregenzandClaude Opus 5 332ae7d183 [eric] agents: Stop disarms every queued injection, so a stopped agent can never resurrect
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014wtspwSFzZmjCx9UNPAorQ
2026-08-20 15:50:55 -07:00
ciregenzandClaude Opus 5 ea04b1da11 [eric] agents: one provider verdict per ask, and a spent plan is never sold as a short wait
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014wtspwSFzZmjCx9UNPAorQ
2026-08-20 15:08:12 -07:00
ciregenzandClaude Opus 5 7c04c755ae [eric] agents: an invisible thinking pill no longer breaks the repeat-card dedup
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014wtspwSFzZmjCx9UNPAorQ
2026-08-20 12:32:14 -07:00
ciregenzandClaude Opus 5 e5e8da37ac [eric] agents: a provider's error never reaches the user as the agent's own words
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014wtspwSFzZmjCx9UNPAorQ
2026-08-20 12:22:34 -07:00
ciregenzandClaude Opus 5 0f2543db65 [eric] agents: a finished turn always leaves the user something to read
Every silent-stop fix so far was a detector for a shape somebody had already
found in the field, which is why the class kept coming back wearing a new hat.
turn_spoke.py moves the question down a tier: at the one exit every terminal
path passes through, ask whether anything readable appeared since the user last
spoke, and if not, say the honest line. Cause no longer has to be enumerated for
the user to be answered.

proactive_prune.py is the hermes trigger we were missing. Their own tests say
our bug out loud: on a large window, a percentage-of-window compaction check
almost never fires, so aged tool output rides in history and is re-sent verbatim
every turn. Measured here, our shaping cut 0.0% at every session size; with a
fixed 60K-token cost trigger it cuts 88% at 12 turns, 93% at 30, 94% at 60. The
prompt-cache contract is load-bearing rather than optional, because our prune is
a rebuild: it commits only when it reclaims enough to pay for the busted prefix,
then disarms until history has regrown a full runway.

Also: lane preflight now treats only 401/403 as a dead credential, since
testStatus=="unavailable" conflated a throttled lane with a revoked one and told
users to reconnect a merely rate-limited Gemini; and awaiting_reconnect is
cleared when the retry budget is spent, so a stale flag can no longer muzzle the
floor and end an ask in total silence.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014wtspwSFzZmjCx9UNPAorQ
2026-08-20 11:36:34 -07:00
ciregenzandClaude Opus 5 9a7570092f [eric] agents: never spend a turn on a lane the router gave up on, and stop our own retries stacking cards
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014wtspwSFzZmjCx9UNPAorQ
2026-08-20 09:20:46 -07:00
ciregenzandClaude Opus 5 58e376496e [eric] agents: an outage parks the turn and resumes when the provider answers; OAuth connects survive a backend restart
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014wtspwSFzZmjCx9UNPAorQ
2026-08-20 08:39:41 -07:00
ciregenzandClaude Opus 5 d5934b1da1 [eric] agents: no turn ends silent, and a codex 401 waits out the rotation window
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014wtspwSFzZmjCx9UNPAorQ
2026-08-20 07:21:00 -07:00
ciregenz 7a4af2c419 [eric] workflows: the restart-loop-guard pause notice called a nonexistent broadcast; it never reached the user (linter catch) 2026-08-19 17:24:23 -07:00
ciregenz a00f277821 [eric] agents: a retry re-failing identically bumps the existing error card instead of stacking clones 2026-08-19 17:24:23 -07:00
ciregenz 4be2f10d02 [eric] agents: every sub lane self-heals a mid-run token expiry before any card; codex retries wait out the rotation window instead of burning the one shot inside it 2026-08-19 17:24:23 -07:00
ciregenz 455b8fa4d3 [eric] agents: MCP responses over the CLI's silent output ceiling get shrunk instead of dropped; a 340KB result line was written and never resolved, the lost-delegation class (ENG-327 root) 2026-08-19 16:43:38 -07:00