dangernoodle-io / ouroboros / 24490004691
81%

DEFAULT BRANCH: main
Ran
Jobs 1
Files 22
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 73.547%. Remained the same
24490004691

push

github

web-flow 
fix(ouroboros-mcp): plugin-qualified skip list, nudges to stderr, tighter agent descriptions (#30) (#36)

three related cleanups to cut token waste and false-positive nudges
on plugin-provided agents:

- nudge/persist/error output from stop, subagent-stop, post-commit-nudge,
  post-edit-check moves from stdout to stderr. all hook stdout is
  injected into claude's next-turn context, so these lines were paying
  ~15+ tokens per firing for output the user never needed to see.
  user-prompt-context and subagent-start keep stdout — their output
  IS the feature (KB context injection)
- SKIP_AGENT_TYPES matching now handles plugin-qualified agent_type
  names via a new isSkippedAgentType helper. claude code reports
  plugin agents as ouroboros-mcp:knowledge-explorer, which never
  matched the bare-name skip list, causing (a) subagent-start to
  inject KB context into agents whose whole job is to read the KB,
  and (b) subagent-stop to fire tier-1/tier-2 persistence nudges on
  their output. built-in agents (Explore, Plan) still match because
  names without a colon fall through
- rewrite knowledge-explorer and backlog-manager descriptions: lead
  with "DEFAULT" routing language, widen trigger examples, drop
  commentary blocks. descriptions are loaded into main context on
  every turn, so the rewrite both improves routing and cuts tokens
  vs the original verbose form

1329 of 1807 relevant lines covered (73.55%)

0.84 hits per line

Jobs
ID Job ID Ran Files Coverage
1 24490004691.1 22
73.55
 GitHub Action Run
Source Files on build 24490004691