Parfii-bot a4e667de10 KeiSeiKit-public — clean state

Single-commit clean baseline after security scrub of niche-tells,
project codenames, internal jargon, and contributor-email leaks.

Contents:
- 100 Rust crates (_primitives/_rust/)
- 37 agent manifests (_manifests/) + generated specs (_generated/)
- 67 user-invocable skills (skills/)
- 33 hooks (hooks/)
- Composition blocks (_blocks/)
- Documentation (docs/, README.md)
- TS adapter packages (_ts_packages/)
- Assembler (_assembler/)
- Roles (_roles/)
- Templates (_templates/)
- Forgejo CI (.forgejo/)

Author: Denis Parfionovich <info@greendragon.info>

License: see LICENSE.

2026-05-01 12:09:03 +08:00

1.9 KiB

Raw Blame History

Phase 1 — Task intake (one free-text field)

The single free-text field in the wizard. Everything else is a click.

1a — Prerequisite check

Before prompting, verify the v0.11 pipeline is live:

# Resolve sync-repo path from env or secrets file.
# shellcheck disable=SC1091
[ -f "${HOME}/.claude/secrets/.env" ] && . "${HOME}/.claude/secrets/.env"
REPO_PATH="${KEI_MEMORY_REPO_PATH:-}"
QUEUE_SH="${HOME}/.claude/agents/_primitives/kei-sleep-queue.sh"

if [ -z "$REPO_PATH" ] || [ ! -d "$REPO_PATH/.git" ] || [ ! -x "$QUEUE_SH" ]; then
    printf 'v0.11 sleep-sync not configured — run `/sleep-setup` first, then retry.\n'
    exit 0
fi

If either check fails, exit the wizard. Do not offer offline queue mode.

1b — Free-text prompt

Emit a plain chat message (NOT AskUserQuestion — a free-text message is fine when it is the only typed field and has a trivial non-empty validator):

What are you sleeping on? One or two sentences — the nightly agent will read this verbatim. Examples:

"Should I pick a continuous-time net or a small transformer as the memory re-ranker?"

"Compare SvelteKit, Astro, and Next.js App Router for the kit's landing page."

"What pattern in recent audit-backlog entries has the highest fix-value-per-effort?"

Store the reply as TASK_TEXT.

1c — Validate

Reject if TASK_TEXT is empty or only whitespace.
Reject if TASK_TEXT > 4000 characters (keep queue files small; the agent has 15 minutes wall-clock anyway — a novella does not help).

On reject, print

Task description must be non-empty and <= 4000 chars. Try again?

and re-prompt. Up to 3 attempts; on the 3rd empty submission abort the wizard with a short "try again with /sleep-on-it" message.

Verify-criterion

TASK_TEXT is non-empty and <= 4000 chars.
The v0.11 sync pipeline is wired (REPO_PATH + QUEUE_SH exist).
Exactly ZERO AskUserQuestion in this phase (free-text message only).

1.9 KiB Raw Blame History