Menu
Use when the user explicitly invokes /chameleon-journey to run the comprehensive real-world journey harness against the chameleon plugin
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
Active when .chameleon/ profile directory exists. Explains hook-injected pattern context and violation feedback for TypeScript/JavaScript, Ruby, and Python as first-class languages — framework-agnostic by default (it learns your repo's own conventions), with deeper awareness where conventions are strong: Rails for Ruby, Django/DRF/Flask/FastAPI for Python, and Next.js/NestJS for TypeScript/JavaScript.
Use when the user explicitly invokes /chameleon-deep-work <task> to execute a substantive coding task end to end: dig the code and external unknowns first (no clarifying questions — parallel expert subagents work independent unknowns), present a 100%-understanding brief, then implement and verify in an isolated git worktree under chameleon's per-edit guardrails.
Use when the user explicitly invokes /chameleon-init to bootstrap a chameleon profile for the current repository (TypeScript/JavaScript, Ruby, or Python — framework-agnostic, with deeper awareness for Rails and Django/DRF/Flask/FastAPI)
Use when the user explicitly invokes /chameleon-trust to approve a committed chameleon profile for use in their current Claude Code session
Use when the user explicitly invokes /chameleon-pr-review to review a PR or branch diff against the repo's chameleon conventions, principles, and task requirements. Reports convention violations + logic gaps.
Use when the user explicitly invokes /chameleon-receiving-code-review to handle a code review the team left on their PR — verify each comment against the code and the repo's chameleon conventions, decide apply-or-push-back, draft replies, and implement approved fixes one at a time.
| name | chameleon-journey |
| argument-hint | [--list | --dry-run | --acts A,B | --model ID | --max-budget-usd N] |
| description | Use when the user explicitly invokes /chameleon-journey to run the comprehensive real-world journey harness against the chameleon plugin |
Run the journey harness at tests/journey/. The harness verifies chameleon's full lifecycle by spawning real claude -p subprocesses against committed seed fixtures.
Requires a development clone of the chameleon repository. The installed plugin ships only the runtime surface (hooks/, skills/, mcp/, ...); it does not include tests/. Every command below runs from the root of the CLONE, not from the installed plugin directory.
Full run: 21 acts, ~$40 typical spend, ~95 min runtime, $100 hard budget cap (default --max-budget-usd).
The cap is deliberately well above the typical spend. It is checked against the sum of the selected acts' per-act CEILINGS (~$80 for all 21), which are upper bounds, not estimates — a cap set near the typical spend aborts a healthy run partway, after the money is gone and with the gate incomplete. Quote the ~$40 typical figure to the user, and the $100 cap as the worst case they are authorizing.
A full run bills real API usage and takes about an hour, so it is not something to start on the
user's behalf from a bare /chameleon-journey. Before the first Claude-spawning command:
--dry-run (free) to confirm the preflight passes, and --list if the user wants to see
what would execute.Skip the confirmation only when the user's own message already authorized the spend (an explicit
"yes, run the full journey" or a --max-budget-usd they chose themselves). A run gated to zero
spend (--dry-run, --list) needs no confirmation.
Only after the user has confirmed. From the chameleon repo (clone) root:
PYTHONPATH=. plugin/mcp/.venv/bin/python -m tests.journey.runner
Variations:
--list: show acts + phase coverage, exit 0.--dry-run: run preflight only (claude on PATH, git >= 2.28, fixtures present, plugin/mcp/.venv), exit before any Claude spawn.--max-budget-usd N: pre-flight + mid-run abort if projected cost exceeds N (default 100). Checked
against the SUM of the selected acts' ceilings, so an N below that sum refuses to start rather
than running partway — pair a small N with --acts.--acts A,B: run only these acts. Dependencies are NOT auto-added: an act reusing a bootstrapped
and trusted fixture needs its provider too (ts_basic comes from 02_init_flow, rails_basic
from 07_rails_parity).--model ID: model each act's worker spawns with (default: the floating sonnet alias). Pin an
exact id such as claude-sonnet-5 when the results will be compared across runs.--results-dir DIR: override per-run output dir (default tests/journey/results/).[ACT N] ... markers + cost + duration.tests/journey/results/journey_<ts>/run.json + run.md: per-act + per-phase results.tests/journey/results/journey_<ts>/checkpoints/<act>.jsonl: per-act checkpoint data for post-mortem.The full run requires:
claude CLI on PATH with API access.git init --initial-branch=main).plugin/mcp/.venv/bin/python (run cd plugin/mcp && uv sync if missing).The harness writes ALL state to a per-run dir; the developer's own ~/.local/share/chameleon/ is never touched.