.zshrc sources agent-env.zsh from there; README points to the new home.
History of the moved files stays here up to e000cfe.
Co-Authored-By: Claude Mythos 5.1 <noreply@anthropic.com>
- Git section carries the rules that lived in auto-memory: push and PRs
without asking, no --amend or stash pop with parallel sessions, absolute
commit-message path, readable PR branches, push unsandboxed
- External communication: no notice on automated texts, terse PR replies,
local time in chat
- Model selection: opus for research and judgement work; rule 0 for the
self-hosted backend, fed by agents/backend-hint.sh as a SessionStart hook
(prints a context line only when ANTHROPIC_BASE_URL is the vLLM gateway)
Co-Authored-By: Claude Mythos 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UH8jPP1zXteddiR9pEd4KE
Identity, write-action confirmation, external communication, error
handling, git commits, Jira defaults and the icinga/isac skill pointers
apply in every repo, not only in ai; the ai AGENTS.md keeps repo-only rules.
Co-Authored-By: Claude Mythos 5.1 <noreply@anthropic.com>
- cognee section 67 → 30 lines: triggers and routing only, how-to lives in
the plugin skill; recall/remember written as `cognee` subcommands
- memory writes explicitly exempt from project write-confirmation rules;
precedence for projects with their own findings convention (file first,
cognee note with link)
- removed-hook JSON dropped (in git history), one-targeted-retry rule added
- ccc gotchas moved to the ccc skill, md paragraph translated and shortened
- AGENTS-claude.md: onyx block no longer duplicates the plugin description
Co-Authored-By: Claude Mythos 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HJuj5BdSLtn9R4xHduxYWY
Der Verweis auf ~/.local/bin/md und das ai-Repo war zum Aufrufzeitpunkt
irrelevant und machte den Eintrag von einem zweiten Repo abhängig.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014Jmx1FL5pssYBKsCjj2pJr
Outline mit Token-Größe pro Abschnitt, dann gezielt den Teilbaum lesen —
statt head oder grep auf Überschriftenzeilen.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014Jmx1FL5pssYBKsCjj2pJr
Deterministische Tools (grep/read_file/get_diff/find_symbol_*) für
Recherche in nicht geklonten Seibert-Repos; ask_codebase nur auf
explizite Anweisung (verschachtelter Server-Agent, Minuten pro Call).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Q6JHuFiJ86AyyMqyTaTcHi
Geprüft: op read "op://Employee/<item>/password" zeigt denselben
generischen full-access-Dialog wie op item get. 1Password autorisiert das
Werkzeug, nie den Zugriff — auch wenn dieser vollständig spezifiziert ist.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012aD7fG6BpN1HchyHeNXZb4
Die imperative Hälfte der Key-Architektur lag unversioniert in ~/.zshrc und
~/.local/bin, während die deklarative (.env.agents) getrackt war. Jetzt liegt
alles Agent-bezogene hier:
- agents/agent-env.zsh — SSH_AUTH_SOCK + agent-env(), gesourced von .zshrc
- agents/ensure-machine-agent — startet den Maschinen-Agent und stellt beide
Keys sicher (ersetzt load-signing-key, das nur den Signing-Key kannte,
während agent-env separat den Agent-Key lud)
- agents/README.md — Design-Doc, verschoben aus dem ai-Repo zum Code
.env.agents bleibt rein deklarativ, weil op run --env-file die Datei nur
parst und keinen Code ausführt — daher die Zweiteilung.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012aD7fG6BpN1HchyHeNXZb4
Die Datei trägt mit GIT_CONFIG_* eine Signing-Config, die nur mit dem
Ein-Key-Agent funktioniert — Konsumenten, die sie ohne agent-env laden
(launchd-Jobs, IDE-Plugins), scheiterten sonst mit 'Couldn't find key
in agent?' am leeren Apple-Agent.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012aD7fG6BpN1HchyHeNXZb4
Agent-Sessions nutzen mwiegand-agent über den geteilten ssh-agent
(~/.ssh/isac-agent.sock, befüllt von agent-env) und signieren Commits
per GIT_CONFIG_* mit dem Agent-Key statt op-ssh-sign.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012aD7fG6BpN1HchyHeNXZb4
Claude Code's Bash tool escapes every ! to \! at the transport layer
(anthropics/claude-code#61121, a regression of a 2.1.87 fix), corrupting
Python !=, jq, Jira JQL, branch names, etc. Documents the single robust
workaround: write !-containing code to .tmp/ via the Write tool and run
the file; .tmp/ is prompt-free via the Edit(.tmp/**) allow rule.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- Replace fixed "5h"/"7d" window labels with hours/days remaining until
the rate-limit window resets, derived from rate_limits.*.resets_at.
- Wire up the previously-captured-but-unused $thinking variable as a
fallback marker when no explicit effort.level is present.
- Color-grade all three percentages (context, 5h, 7d): yellow >=80%,
red >=90%. Uses \e[39m to restore default fg without losing the
surrounding dim attribute.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
