repo-recon
Skill0 starsChanged 3 months agoMITView on GitHub
Parallel read-only recon of local git projects via subagents for survey, deep-dive, or pre-enhancement analysis.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- JPeetz/agent-skillstheir own version, first seen Nov 14, 2026
- gonun13/agent-skillstheir own version, first seen Sep 5, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- JPeetz/agent-skills★ 17same name
- gonun13/agent-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository4 other files it gives its agents.
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 6 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 19
No subagent facility: fall back to sequential reads and shrink scope (fewer projects, tighter budgets).
- Distinctive: line 28
If targets are ambiguous, present the candidate list and ask.
- Distinctive: line 32
Default 3 projects per subagent.
- Distinctive: line 33
Default ~15 min reading budget per project.
- Very specific: line 35
Write each project's quick-scan report to <project>/dsh-scan/recon-scan-<project>.md — inside the analyzed project, never a CWD-relative …
- Distinctive: line 36
After completion: merge into one summary for the user (one-liner per project), then ask which projects to deep-dive
- Distinctive: line 51
Always git pull before deep dive.
- Very specific: line 54
All reports live inside the project they describe (<project>/dsh-scan/).
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Repo Recon
Two-phase local project reconnaissance with subagents. Phase 1 = quick scan (breadth). Phase 2 = deep dive (depth, only after user confirms targets). Never start the enhancement/analysis phase without explicit user confirmation.
Parallel-subagent primitive (host-agnostic)
Both phases delegate reading to subagents. The primitive is: spawn all independent readers in ONE parallel block, then wait push-style — never poll status. Host mapping:
- Codex:
sessions_spawn(mode=run,cleanup=delete), then onesessions_yield; never loop onsessions_list/sessions_history - Other harness hosts: the host's parallel subagent facility — same shape: one spawn block, push-based completion
- No subagent facility: fall back to sequential reads and shrink scope (fewer projects, tighter budgets); say so in the report
Every subagent brief must state: READ-ONLY (never modify project files; the only permitted write is that project's own report under <project>/dsh-scan/); method = README + PLAN/IMPLEMENTATION/AGENTS docs first, then main-entry source only; a reading budget; and that long output goes to report files because subagent text replies truncate.
Phase 0 — Triage (self, no subagent)
- List candidate projects from the roots the user named; if none were named, probe conventional project roots (e.g.
~/git/*/,~/src/*/,~/projects/*/) and say which roots you scanned - Skim each README (first ~400 chars; if none, list top-level files)
- Filter by the user's stated criteria (example: "projects with LLM orchestration"). Skip infra/tooling repos unless asked
- If targets are ambiguous, present the candidate list and ask; otherwise proceed to Phase 1
Phase 1 — Quick scan (parallel subagents)
- Default 3 projects per subagent; adjust group size to the project count
- Default ~15 min reading budget per project; state the budget in the brief
- Per-project report format (≤40 lines): 定位 / 技术栈 / 工作流拆解 / AI 编排点 / 编排痛点
- Write each project's quick-scan report to
<project>/dsh-scan/recon-scan-<project>.md— inside the analyzed project, never a CWD-relative directory; createdsh-scan/if missing - After completion: merge into one summary for the user (one-liner per project), then ask which projects to deep-dive
Phase 2 — Deep dive (one subagent per confirmed target)
git pull --ff-onlyeach target first; report branch/HEAD and new commits to the user- One subagent per project, spawned in one parallel block; default ~40 min reading budget
- Feed phase-1 findings into the brief as "claims to verify and expand"
- Deep-dive brief: every claim needs file:line evidence; sections = HEAD state / full workflow map / LLM call graph / state & cache / error handling (retry/timeout/降级 inventory) / deployment / pain-point list (each mapped to what plugin-style orchestration would fix + severity 高/中/低)
- "未深入" annotation beats guessing; collect them into an explicit list
- Report to
<project>/dsh-scan/recon-deep-<project>.md(inside the analyzed project; ≤300 lines, 中文); final subagent reply = 1-line workflow summary + Top-5 pain points - Summarize both reports to the user and wait for confirmation before writing any enhancement analysis
Rules
- GB-size repos: README + docs + core pipeline files only; never full-tree reads
- Always
git pullbefore deep dive; if pull brings new commits, tell the user what changed - Push-based completion only: yield and wait; never poll
- Deep-dive subagents get read-only scope plus exactly one writable report path each (under the target's
dsh-scan/) - All reports live inside the project they describe (
<project>/dsh-scan/); never write suite artifacts to the current working directory or home.dsh-scan/stays untracked — the user decides whether to commit or ignore it - Record the project path + branch + HEAD in every report header
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

