promotion-assistantlisted

# promotion-assistant > Governing principle (full text in the repo's `PHILOSOPHY.md`): **methodology is constant, signals > adapt; compliance is engineering not goodwill; dry-run is the default, not an option.** The channel > matrix, six-layer funnel, bandit and compliance gate are fixed; every platform limit, audience and > piece of copy lives in per-product config. No outbound action ever leaves the machine unless the > product is explicitly set live AND that channel is individually authorized. ## When to use / when to stop - **Use** when promoting a *shipped* product across many channels with quantified feedback: bulk email + multi-platform posting (覆盖式) and forum replies + DMs (精准式), with daily multi-account upkeep, funnel attribution, and self-evolving tactics. - **Stop / route elsewhere:** a one-off single post → just post it. Pure market/competitor research → `market-intel`. Scheduling/reminders themselves → this skill *delegates* to the `schedule-reminder` base (it does not reimplement scheduling). ## Architecture (six layers, two repos, load `reference/<shard>.md` on demand) | layer | job | shard | |---|---|---| | L0 config/creds | locate per-product config (`PROMO_CONFIG_DIR`, Mode B secrets) | `reference/config-schema.md` | | L1 orchestration | plan→`schedule-reminder`; run `--once/--daemon`; alerts→Discord relay | `reference/integration.md` | | L2 channels | one provider/platform (`publish/engage/dm/read_metrics`) | `reference/channels.md` | | L3 comp