wassname/ml_debug
1
0
Fork 0
mirror of https://github.com/wassname/ml_debug.git synced 2026-08-25 11:21:20 +08:00
Code Issues Packages Projects Releases Wiki Activity

readme: match the skill's wording, instructions not rituals

Browse Source 
Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
This commit is contained in:
wassname
2026-08-19 20:46:48 +08:00
co-authored by Claudypoo
parent 8b092a4320
commit c858469f59
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+1 -1
View File
@@ -14,7 +14,7 @@ Or paste `SKILL.md` into your system prompt / context when debugging.
- **This README** -- the folklore, for humans: verbatim sourced quotes from practitioners, general lessons first, modern transformers and LLM fine-tuning in their own section.
- **[SKILL.md](SKILL.md)** -- what an agent loads: the rituals the folklore implies, each with a trigger, a form to fill, and an artifact to show the user. "Assume you have a bug" becomes "send a subagent to find one and report what it found". This is a bet that a form gets filled where a principle gets skipped, and it is untested. The bet is worth making because the folklore version measured no gain (below), and because forms have their own failure mode: they get filled with plausible content that nobody checked.
- **[SKILL.md](SKILL.md)** -- what an agent loads: the folklore turned into instructions, each with a trigger, a form to fill, and an artifact to show the user. "Assume you have a bug" becomes "send a subagent to find one and report what it found". This is a bet that a form gets filled where a principle gets skipped, and it is untested. The bet is worth making because the folklore version measured no gain (below), and because forms have their own failure mode: they get filled with plausible content that nobody checked.
- **[PLAYBOOK.md](PLAYBOOK.md)** -- the synthesized long-form: mental models, practitioner priors, step catalogs, symptom tables, the agent debugging loop, triage, and anti-patterns. Menus of hypotheses distilled from the same sources, not quotes. Deeper one-off tricks (loss-surface analysis, stuck-metric diagnosis, sweep reliability) live in [refs/](refs/).