fable-lead

Plugin fable-lead from marketplace fable-lead in oZeppy/fable-lead-plugin. Commands read from the repo's manifest — 2 days ago.

Skills (1)

• fable-lead

  Bootstrap the Fable-led working system into the current session and project: verify/install the global working-system files (user CLAUDE.md + scout/finder/builder/builder-tdd/verifier/adjudicator/synthesizer agents), scaffold the project's CLAUDE.md and PROJECT_MEMORY if absent, confirm model/effort

Subagents (7)

• adjudicator

  Vote-split adjudicator. Use ONLY when two independent verifier votes on the same claim disagree (v1 != v2): disagreement is where the uncertainty is, so it is the one place the strongest configuration is spent. Rare by design (about one claim in fifty) — not a general-purpose senior reviewer. <!-- B

• builder-tdd

  Test-driven implementation agent: use when FAILING TESTS already define "done" — a bug with a reproducing test, a feature whose acceptance tests exist. If no test defines done, write the tests first or use fable-lead:builder. <!-- BEGIN pins:runs -->Runs claude-sonnet-5-5/high by exact id (006 C, 12

• builder

  Implementation agent: build or fix one well-scoped change end to end — an adapter, a bug fix, a migration, a probe — and prove it works. Use this INSTEAD of general-purpose for delegated implementation: general-purpose inherits the session's effort, and unpinned builders were the largest single line

• finder

  Coverage agent for a pipeline: sweep a surface and RETURN A SET — findings, a dataset, measured rates, a map of code paths, harvested sources — that later stages will verify, rank or build on. A verifier downstream catches its FALSE claims but nothing can restore what it MISSES, so it runs the confi

• scout

  The head's reading delegate. Use for any read likely to exceed ~150 lines, a search across more than a few files, a log or transcript trawl, or a long command output: it locates, reads and summarises, writes the detail to a file, and returns a short answer with file:line pointers. Cheap on purpose —

• synthesizer

  Head-class synthesis agent: the ONE subagent per workflow whose output ships verbatim — final reports, briefs, decision memos that integrate many agents' findings. Reserve for synthesis and for verification that turns on novel reasoning rather than more evidence; everything else goes to finder or ve

• verifier

  Adversarial verifier. Use to REFUTE a claim, finding, statistic, or diff before it ships: re-derive numbers from source, re-run splits, fetch cited pages, probe edge cases. Its misses ship; rigor comes from a SECOND independent vote on ship-critical claims, not from effort. Use PROACTIVELY after sub

Hooks (3)

• PreToolUse

  python3 "${CLAUDE_PLUGIN_ROOT}/bin/fable-lead-bash-guard" --hook 2>/dev/null || true

• PostToolUse

  python3 "${CLAUDE_PLUGIN_ROOT}/bin/fable-lead-fallback-check" --hook 2>/dev/null || true

• UserPromptSubmit

  python3 "${CLAUDE_PLUGIN_ROOT}/bin/fable-lead-fallback-check" --hook 2>/dev/null || true

Similar plugins