Hacktoberfest 2026: le issue che i maintainer hanno segnato per ottobre, aperte e adatte ai principianti. Sfoglia le issue Hacktoberfest

[Feature]: skill.md — "Do not skip steps" contradicts the documented skips; shouted MANDATORY Agent-tool line; API-key note stated twice

Aperta
  • Qualcosa non va?
#4,061 2 commenti 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 1 giorno

@venkatpachala ci sta già lavorando.

Dal 5/10/2026.

  • #4110 di @venkatpachala — aperta

Valutazione

Difficoltà
3/5
Tempo stimato
1-2 giorni
Idoneità per principianti
74/100
Tipo di issue
Documentazione
Chiarezza
Specificata chiaramente
Stato di attività
Attiva
Stack tecnologico
python
Ambito
developer-experience, documentation

Direzione di ricerca

Start in tools/skillgen/fragments/core/core.md and review related issue #2513 for the API-key wording. Run uv run python -m tools.skillgen --bless, then inspect the generated graphify/skill*.md files. Done means the skip instructions and Agent-tool guidance no longer conflict, the API-key note appears once, and generated skills reflect core.md.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

good first issue help wanted
Problem or use case

graphify/skill.md (generated from tools/skillgen/fragments/core/core.md) gives the agent instructions that conflict:

  1. "Follow these steps in order. Do not skip steps." (skill.md:59, core.md:56) follows the fast path at :53, which says to skip Steps 1–5 entirely when graphify-out/graph.json exists and the request is a question. The steps then skip conditionally themselves (Step 0 only for a URL or several paths — "A plain local path skips this step"; Part B has its own "skip Part B entirely" fast path). An agent that follows instructions literally must break one of them; "do not skip" is the absolute one, so the cost is a needless rebuild instead of a graphify query.
  2. :214 — "MANDATORY: You MUST use the Agent tool here. Reading files yourself one-by-one is forbidden … If you do not use the Agent tool you are doing this wrong." Current Claude models follow instructions closely and tend to over-apply stacked caps/absolutes — here, spawning subagents even when the code-only fast path or a tiny doc set makes Part B unnecessary. The reason (5–10× faster) is the load-bearing part.
  3. API-key note stated twice — :163 and :170 say the same thing in two blockquotes, and #2513 reports the claim itself is inaccurate (llm.py detect_backend()). Whatever #2513 settles should appear once.
Proposed solution

In tools/skillgen/fragments/core/core.md, then uv run python -m tools.skillgen --bless:

  1. Replace "Follow these steps in order. Do not skip steps." with: "Run the steps in order. Skip a step only where it says so (the fast path above, Step 0 for a plain local path, Part B for a code-only corpus)."
  2. Replace the MANDATORY line with: "Dispatch semantic extraction with the Agent tool — one subagent per chunk; it is 5–10× faster than reading files yourself. Skip this when the Part B fast path applies."
  3. Merge the two API-key blockquotes into one, using whatever wording #2513 lands on.
Alternatives considered

Fix only the contradiction in (1) and keep the absolute wording in (2) — resolves the conflict but keeps the over-application risk. Splitting the skill (#1641) is complementary, not a substitute.

Area

Skills or assistant integrations

Are you willing to submit a PR?

No, I just wanted to request it

Compatibility / behavior impact

Generated skills only (every graphify/skill*.md built from core.md changes on regenerate). No effect on node/edge IDs, cache semantics, graph.json schema, MCP output, or incremental updates.

Additional context

Related: #2513, #1641, #3768. Found in 0.9.53; confirmed unchanged on v8 HEAD (core.md:50/56/167).

Lingua principale
Python
Stelle
124k
Fork
11.9k
Nessuna PR unita negli ultimi 30g

Preparare l'ambiente

Questo progetto non fornisce container di sviluppo, Dockerfile né guida per i contributori, quindi l'ambiente è a tuo carico: parti dal suo README e consulta la nostra guida al primo contributo per i passaggi generali.

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di Graphify-Labs/graphify

Tutte le issue di Graphify-Labs/graphify

Issue simili

Altre issue su Python

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.