nexus
Skill0 starsChanged 2 months agoMITView on GitHub
Map an unfamiliar repository, select task-relevant implementation and verification paths, and build a code-cited context pack for planning, debugging, review, or handoff.
- Reads credentials
10 other projects use a version of this.
Adopted by
Most recently taken up by
- simota/agent-skillstheir own version, first seen Oct 23, 2026
- Eskyee/agentbot-opensourcetheir own version, first seen Oct 13, 2026
- ProfSynapse/nexustheir own version, first seen Oct 9, 2026
- BankrBot/openclaw-skillstheir own version, first seen Oct 8, 2026
- aayushostwal/nexustheir own version, first seen Aug 10, 2026
- GPire/claude-skills-swarmtheir own version, first seen Jun 30, 2026
Used with
Projects with it also use aeon-autoresearch, aeon-deal-flow and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
BankrBotSkillaeon-autoresearch
Evolve any installed skill by generating four variations along separate theses (better inputs / sharper output / more robust / rethink), scoring them on a weighted rubric, and applying the winner. Never downgrades a working skill — aborts cleanly if no variation improves the original. Use when an installed skill is producing low-signal output, hitting deprecated APIs, or feels stale. Triggers: "improve this skill", "evolve $skill", "auto-research my X", "regenerate variations".
Also in 3 projects that use this
BankrBotSkillaeon-deal-flow
Weekly funding round tracker across configurable verticals (AI, crypto, prediction markets, agentic infrastructure, etc.). Primary-source-required (Form D, press release, or verified founder/firm post) — filters re-announcements and rumored rounds. Per round: lead, size, valuation if disclosed, what they do, why now, and the sharpest risk. Triggers: "this week in funding", "deal flow", "who raised this week", "crypto funding rounds", "AI funding tracker".
Also in 3 projects that use this
- +10 more from BankrBot
Other versions10 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 10
- BankrBot/openclaw-skills★ 1.2ksame name
- BankrBot/skills★ 1.2ksame name
- ProfSynapse/nexus★ 158same name
- simota/agent-skills★ 96same name
- aayushostwal/nexus★ 19same name
- Eskyee/agentbot-opensource★ 17same name
- GPire/claude-skills-swarm★ 8same name
- Aether-liusiqi/nexus★ 2same name
- InfiniteRoomLabs/agent-ops★ 0same name
- terrene-foundation/pact★ 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.
The file
We split this skill into the rules it gives an agent
5 rules. 0 are common, 3 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 8 edited versions we read keep these.
3
Distinctive
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 13
Read applicable repository instructions before inspecting scoped files.
- Very specific: line 18
Run the bundled scripts/map_directory.py using its installed absolute location, the authorized root, and a new output path:
- Distinctive: line 41
Read the smallest coherent set answering the questions: authoritative overview/manifest → entry point/caller → implementation → …
- Distinctive: line 45
Stop expanding when questions are answered and the execution path is coherent.
- Very specific: line 49
Read the context-pack template when composing the handoff.
“Common” is measured: kept by at least 6 in 10 of the 8 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Nexus
Reconstruct a repository in two passes: map broadly, then read selectively. Connect implementation, contracts, configuration, tests, and operational paths. A map is an inventory, not semantic understanding.
Establish the task and snapshot
- State the outcome, constraints, and non-goals. Resolve the explicit repository root or use
git rev-parse --show-toplevel; label a non-Git directory honestly. - Read applicable repository instructions before inspecting scoped files. Record
git rev-parse HEADandgit status --shortwhen available. Preserve dirty changes. - Choose a context directory outside the repository by default. Treat its contents as private working material; source names alone can be sensitive. Sharing a portable pack requires an explicit disclosure boundary.
Map broadly
Run the bundled scripts/map_directory.py using its installed absolute location, the authorized root, and a new output path:
python3 <installed-skill>/scripts/map_directory.py <repository-root> <context-directory>/directory-map.mdThe Python 3.10+ standard-library mapper lists paths and top-level Python functions/classes. It skips common generated/cache directories, hidden entries and sensitive filename patterns by default, never follows symlinks, bounds Python source reads, and refuses to overwrite an output. New map files use owner-only POSIX permissions. It parses Python but never executes repository code. Review --help to add task-specific exclusions with --ignore-dir and --ignore-file. Exclusions are risk reduction, not a secret detector: review the result before disclosure. The map omits function bodies, values, and string literals; even names may need redaction.
If mapping fails, report the exact failure. Do not silently present another tool's inventory as this mapper's output. Partial parsing failures are annotated without copying source or exception contents. Read the map entirely when practical; otherwise use line counts and chunked reads. Identify entry points, manifests, schemas, migrations, tests, docs, CI, deployment surfaces, and generated-code boundaries.
Select zoom areas
Translate the task into a small set of investigation questions. Search the map and scoped repository using rg for task nouns, symbols, routes, configuration keys, errors, and tests. Select an area only when it:
- implements or invokes the target behavior;
- defines an interface, model, schema, policy, or configuration contract;
- verifies behavior in tests or operational checks; or
- controls a relevant build, migration, delivery, or runtime path.
For each area record why it matters, files/symbols, the question it answers, and its relation to other areas. Include implementation and verification paths when they exist. Do not broaden into adjacent subsystems just because they are interesting.
Zoom in and trace behavior
Read the smallest coherent set answering the questions: authoritative overview/manifest → entry point/caller → implementation → interfaces/configuration → tests/fixtures → relevant operational path. Trace imports, calls, transformations, state changes, and configuration across boundaries.
Use line-numbered reads and cite repository-relative paths plus line ranges. Treat comments/docs as claims until corroborated. Do not read or reproduce secrets, credential stores, private keys, .env values, dependency caches, or unrelated generated artifacts. Do not execute unfamiliar repository code merely to map it. Tests or other executions need their own scope and side-effect assessment.
Stop expanding when questions are answered and the execution path is coherent; retain contradictions and unknowns. A static trace does not establish runtime correctness.
Build and verify the context pack
Read the context-pack template when composing the handoff. Write context-pack.md, directory-map.md, and sources.txt (ordered evidence paths, one per line) together at the chosen destination.
Separate observed facts, reasoned inferences, and open questions. Cite important claims, record revision and dirty-worktree caveat, explain only task-relevant architecture, name the likely change surface without claiming edits, derive validation commands from repository evidence, and explain exclusions. Quote only decisive snippets, not entire files.
Before handoff, verify that cited paths/lines still resolve at the recorded snapshot; the call/data flow is consistent; implementation, contracts, and verification are represented where available; the evidence manifest matches actual reads; and no secrets or unrelated private material entered the pack. Report map/context paths, selected areas, unknowns, and one recommended next action. Mapping is read-only apart from the requested local artifacts; it authorizes neither changes nor publication.
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.

