YujunZhou/tellonce / codex
tellonce
VerifiedUse to view, explain, correct, delete or reorganize the user's saved preferences, or to record one by hand when automatic recording is off; learning and rule injection otherwise run automatically through Codex hooks.
docs/README.md
View on GitHub# Codex Tellonce
Codex-native port of the corrected Claude Code `tellonce`.
Since 1.7.1, background automatic recording is enabled, using the current
platform CLI model on redacted user input. `memory_upsert.py disable-hooks` opts
out; existing explicit opt-out is retained. Hard blocking remains opt-in.
## User Flow
The external UX should stay simple:
```bash
git clone <repo> ~/.codex/skills/tellonce
cd /path/to/project
bash ~/.codex/skills/tellonce/codex/install.sh
bash ~/.codex/skills/tellonce/codex/doctor.sh
```
> The Codex installer lives under `codex/` — the repo-root `install.sh` is the
> Claude Code variant and would register Claude Code hooks instead.
**One-time hook trust approval (required).** Codex trusts hooks by a hash of
their exact definition and **silently skips** untrusted ones — no error, no
log. After install (and after ANY later change to the hook definitions,
including re-running `install.sh`, which rewrites/re-orders them), start an
interactive Codex session and approve the tellonce hook entries when Codex
prompts for review. Until then the hooks do not fire at all. For one-off
non-interactive verification: `codex exec --dangerously-bypass-hook-trust ...`.
Internally the package records a project-local audit ledger and uses wrapper-based verification where possible. Users do not need to choose internal modes during normal install.
## Modes
- `audit_only`: default. Records scans and warnings; does not claim hard enforcement.
- `wrapper`: checks output produced through `tellonce_codex exec`.
- `blocking`: opt-in hard-block layer on PostToolUse (ships with no built-in
rules, so on its own it blocks nothing until you add rules).
(The top-level README's `observe → enforce → full` naming maps to
`audit_only → blocking` here; Codex has no shadow-judge mode yet.)
## Current V1 Capabilities
- Project registration under `.codex/tellonce/`.
- `mode.json` as mode authority.
- Append-only `events.jsonl` with centralized redaction.
- Scan events for preference/pitfall/friction/none.
- Explicit memory promotion with intent and commit events.
- Active memory index rebuilt from committed memory.
- Doctor status line and private path audit.
- Conservative install/uninstall.
- Wrapper run capture and verifier verdict.
- Migration preview that does not write active memory.
- Skill package wrapper scripts.
## Non-Guarantees
- Native interactive Codex text is advisory unless routed through the wrapper.
- Plain `codex`, IDE integrations, and web Codex can bypass wrapper checks.
- Hard blocking is not default.
- Shadow judge is not enabled by default and must not require Claude CLI.
## Claude Code Parity
See `CC_PARITY_MATRIX.md`. No Claude Code capability should be dropped; features may be staged internally but must keep a parity row and a replacement path.
## Uninstall
```bash
# Per-project disable only (keeps the global runtime + ~/.codex/hooks.json entries
# so other projects keep working):
bash ~/.codex/skills/tellonce/codex/uninstall.sh
# FULL uninstall — also removes the hook registrations from ~/.codex/hooks.json
# (so the hooks stop firing everywhere) and the global runtime:
bash ~/.codex/skills/tellonce/codex/uninstall.sh --purge-hooks --purge-skill
```
> Note: the hooks keep firing as long as they're registered in
> `~/.codex/hooks.json`. The default uninstall intentionally keeps that
> registration (the global runtime is shared across projects); pass
> `--purge-hooks` to remove it and fully stop the hooks.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |