一键导入
session-begin
Pre-flight checklist for work sessions — loads context, checks hook state, surfaces warnings, and gates on acknowledgment before work begins.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Pre-flight checklist for work sessions — loads context, checks hook state, surfaces warnings, and gates on acknowledgment before work begins.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Dual-lens repo analysis: Creator View (knowledge, insights, home-repo comparison) + Engineer View (health, security, process). Two user-invokable depths (Standard / Deep); Quick Scan is triage-only. Link mining for curated lists. Fit separation via dual scoring lenses. Outputs to .research/analysis/<repo-slug>/.
Structured discovery-first planning for complex tasks. Ask exhaustive categorized questions, build a standalone decision record, then produce a step-by-step implementation plan with audit checkpoints and convergence-loop verification of diagnosis and plan claims for user approval before any code is written.
Interactive behavioral quality audit for individual skills. Evaluates against 12 quality categories (including T25 convergence loop and completion verification) to surface attention management issues and produce actionable decisions. Includes self-audit verification that all decisions were actually implemented. Uses convergence loops in its own discovery phase. Produces a decision record and updated skill files.
Structured workflow for creating or updating skills — from discovery through validation and audit. Produces structurally correct, behaviorally effective skill packages (SKILL.md + companions) using deep-plan-style discovery with skill-specific question categories and convergence-loop verification of codebase claims.
Audit + refresh the JASON-OS / SoNash label catalogs. Dispatches primary + secondary derivation agents in byte-weighted batches, cross-checks their output, and presents drift / low-confidence / disagreement findings for conversational resolution (D12a). Supplementary skill per D4 + D19 — catalogs remain correct via the PostToolUse hook (§S3) + pre-commit validator (§S6) even if this skill is never invoked.
Multi-agent research engine that decomposes questions, dispatches parallel searcher agents, synthesizes findings with citations and confidence levels, runs mandatory contrarian/OTB challenges, gap-pursuit verification, and cross-model verification via Gemini CLI, and produces structured output with downstream adapters, research index, and management commands.
| name | session-begin |
| description | Pre-flight checklist for work sessions — loads context, checks hook state, surfaces warnings, and gates on acknowledgment before work begins. |
| compatibility | agentskills-v1 |
| metadata | {"version":2.1} |
Pre-flight checklist that orients the session: loads context, validates environment, surfaces warnings, and hands off to the user with a goal-selection prompt.
Time budget: Session-begin SHOULD complete in under 3 minutes. If script issues or staleness require extended remediation, present findings and defer fixes to user decision.
JASON-OS v0.1 scope note: This skill was ported from SoNash and trimmed for bootstrap. Sections referencing infrastructure JASON-OS doesn't have yet (health scripts, override/health-score logs, technical-debt index, consolidation pipeline, secrets-decryption gate) are marked DEFERRED inline with a pointer to re-wire them when the corresponding systems land. See BOOTSTRAP_DEFERRED.md for the broader deferral ledger.
Check all three:
If ANY true: "Session #N already active. What would you like to work on?" If ALL false: Proceed with full checklist.
/session-begin/session-end/checkpointThe SessionStart hook in JASON-OS v0.1 runs only check-mcp-servers.js
(and compact-restore.js after compactions). Consolidation, cross-session
validation, dependency install, and build are NOT handled by the hook yet —
when those hooks land, update this section and remove the corresponding
manual steps below.
"Starting Session #N pre-flight on [branch]. Will load context, check hook state, and surface any warnings."
JASON-OS bootstrap has no scripts/secrets/ infrastructure. When an
.env.local.encrypted + decrypt pipeline lands, restore the SoNash-style gate
here (passphrase-via-stdin, never logged).
SessionStart hook does not yet do cross-session validation. When that hook
work lands, restore the prior-session-missed-/session-end check here.
(ROADMAP.md read is DEFERRED — JASON-OS v0.1 does not have ROADMAP.md yet. Next Session Goals live in SESSION_CONTEXT.md for now.)
Report the current branch from git branch --show-current in the session
announcement. If SESSION_CONTEXT.md documents a different branch (in Quick
Status or Next Session Goals), warn the user before proceeding.
git log --oneline -15
Compare recent commits against SESSION_CONTEXT.md's "Next Session Goals" and "Quick Status." If discrepancies found, present them: "Docs appear stale: [specifics]. Update now or defer?"
Let the user decide. Do not auto-update.
Session-gap detection, consolidation status, and prior-research surface all
depend on infrastructure not yet in JASON-OS (npm run session:gaps,
scripts/run-consolidation.js, .research/research-index.jsonl). Restore
when those land.
JASON-OS bootstrap does not yet wire any npm run health scripts. When any of
patterns:check, review:check, lessons:surface, session:gaps,
roadmap:hygiene, reviews:sync, reviews:check-archive, reviews:archive,
hooks:analytics, crossdoc:check, or the velocity/task-dependency reports
land, restore the SoNash-style block here and flip the script-failure-
escalation rule (Critical Rule 4) from vacuous to live.
Until then: note "Health scripts: not yet wired (JASON-OS v0.1)" in the summary.
Check .claude/state/hook-warnings-log.jsonl. If 10+ entries in last 7 days
→ warn. If fewer, skip silently.
(SoNash override-trend and health-score-drop sub-checks are DEFERRED —
.claude/override-log.jsonl and .claude/state/health-score-log.jsonl do
not exist in JASON-OS v0.1. Restore when those logs land.)
No .claude/hook-warnings.json or scripts/sync-warnings-ack.js in JASON-OS
v0.1. When the warning-acknowledgment pipeline lands, restore the gate here.
No .claude/state/session-start-failures.json, pending-test-registry.json,
or decision-documentation hook-warnings entries yet. Restore when those
systems land.
No docs/technical-debt/INDEX.md in JASON-OS v0.1. Restore when a tech-debt
index lands (or adapt to a JASON-OS-native debt surface).
Present using this format:
Session #N — Pre-Flight Summary
Branch: [branch]
Working tree: [clean / N uncommitted]
Health scripts: not yet wired (JASON-OS v0.1)
Hook warnings: [none / N new entries in last 7d]
Deferred infrastructure: see BOOTSTRAP_DEFERRED.md
Next Goals (from SESSION_CONTEXT.md):
[list ALL goals from the "Next Session Goals" section —
do not truncate or cap the list]
After the summary: "Which goal would you like to focus on, or something else?" Reference the surfaced goals — do not use a generic open-ended prompt.
"Session #N pre-flight complete. Ready to work."
Done when: Session counter incremented, context loaded, hook state checked, goal selected.
session-end (receives session context), checkpoint
(mid-session state)| Version | Date | Description |
|---|---|---|
| 2.1 | 2026-04-17 | JASON-OS bootstrap trim: DEFERRED markers for unwired infra |
| 2.0 | 2026-03-16 | Skill-audit rewrite: 31 decisions, 51→73 score |
| 1.0 | 2026-02-25 | Initial implementation |