vc-plan-discovery
GitHub用于发现与当前任务相关的计划文档。通过脚本扫描同功能全深度、其他功能活跃及通用计划,提取YAML元数据并分类展示,为智能体提供完整上下文,支持主动研究与执行阶段调用。
Trigger Scenarios
Install
npx skills add withkynam/vibecode-pro-max-kit --skill vc-plan-discovery -g -y
SKILL.md
Frontmatter
{
"name": "vc-plan-discovery",
"layer": "contract",
"metadata": {
"author": "vibecode-pro-max-kit",
"version": "1.1.0"
},
"description": "Discover related plans for the current task: same feature folder full depth, other features active-only, general-plans active. Like vc-context-discovery but for plan artifacts.",
"argument-hint": "[feature folder name or task description]",
"trigger_keywords": "related plans, what was tried, plan history, feature backlog, plan discovery"
}
vc-plan-discovery
Output style: Follow
process/development-protocols/communication-standards.md— answer-first, plain language, no unexplained jargon, TL;DR on long responses.
Invocation
Primary method — run the auto-discovery script. It applies the Scope Rules below
(same-feature full depth; other features active-only; general-plans active always) and
extracts ONLY the leading YAML frontmatter block of each .md file (no whole-file reads):
node .claude/skills/vc-plan-discovery/scripts/discover-plans.mjs [--feature <name>] [--json]
It groups output into Active Plans / Backlog / Completed / Reports / References (each line:
- [name]: description (path)) and prints a trailing Found N active, N backlog, N completed, N reports, N references line. It never throws on a missing root and exits 0
unless given a bad flag. Use --json for a machine-readable object. Prefer this over the
manual scan below — it is deterministic and avoids loading huge files into context.
Per task-folder artefact colocation, the script scans one level into each
{slug}_{date}/ task folder and surfaces co-located _PLAN_/_REPORT_/_REF_ artefacts;
the sibling reports//references/ dirs are deprecated and only hold legacy artefacts.
The manual scan in Scope Rules below is the FALLBACK for when the script fails.
Purpose
Surface all plans relevant to the current task so agents have full plan context — what was tried, what is in progress, what is deferred, and what reports and references exist — before doing any phase work.
Scope Rules
- Same feature folder (from task context or argument): read ALL of
active/,backlog/,completed/, plus any legacy siblingreports/,references/— surface every file with frontmatter - Other feature folders: read
active/only — surface plans whose frontmatterdescriptionorfeaturefield matches the task domain general-plans/active/: always scangeneral-plans/completed/, legacyreports/,references/: scan only when same-feature folder is not identified
Per task-folder artefact colocation, expect every current artefact (plan, spec, reports, references) INSIDE its {slug}_{date}/ task folder — scan one level into each task folder. The sibling reports//references/ dirs are deprecated and only hold legacy artefacts.
Frontmatter Reading
Read frontmatter fields: name, description, type, feature, phase
Route by description field for relevance matching (same approach as vc-context-discovery).
Skip files without frontmatter or with incomplete frontmatter — log as "no frontmatter, skipped".
Output: grouped list by folder — Active Plans / Backlog / Completed / Reports / References — with name + description per file.
When To Invoke
- First action alongside
vc-context-discoveryat the start of every loop step (research / validate / execute / update-process) - Any time an agent needs to know: what plans exist for this feature, what was tried before, what is deferred, what references exist
Output Format
### Active Plans
- [name]: description (path)
### Backlog
- [name]: description (path)
### Completed
- [name]: description (path)
### Reports
- [name]: description (path)
### References
- [name]: description (path)
Found N active, N backlog, N completed, N reports, N references
Version History
- 3bcb2f9 Current 2026-07-05 14:35


