SirTheOracle

forge-orchestrator

tmux-based orchestrator for multi-agent building. Coordinates work across Claude Code and two Codex workers via forge-bridge. Translates user requests into dispatched tasks with structured audit logging. Replaces forge-dispatch, forge-state.yml, and stage-routing-map.yml.

forge-fix-runner

Drain the GitHub `forge-fix` issue queue through the forge fix pipeline, one service at a time. Tallies actionable issues worst-first, scouts a service into a routing plan (clusters + quick/full tiers), gets human approval, then dispatches fixes (quick = dispatched fix-code; full = forge-pipeline), gating every `Closes #N` on per-issue verification. Trigger when the user wants to fix triaged QA issues, "work the forge-fix queue", or "run the fix runner".

forge-coder

Procedural coding skill. Reads implementation.md, validates diffs against the codebase, applies changes in execution order, runs tests after each commit group, and produces a coder-report.md artifact.

fix-reproducer

Optional first stage of the fix-pipeline. Verifies a reported bug is reproducible and captures deterministic repro steps, environment state, and evidence. Single procedural executor — does not spawn sub-agents and does not diagnose. Branches by environment: dev path runs the bug locally; prod path gathers observational evidence from the deployed environment. Always writes repro.md, even on NOT REPRODUCED. Trigger when the user invokes a fix-pipeline with the reproduce step enabled.

fix-plan-reviewer

Single-agent adversarial reviewer for fix plans. Reads diagnosis.md and fix-plan.md, verifies the plan actually addresses the diagnosed cause, attacks the plan from multiple angles (cause coverage, defense adequacy, scope correctness, regression risk, rollback integrity, test plan integrity, risk honesty), and produces fix-review.md with severity-rated issues. Does NOT propose fixes for the issues — that is the revise mode of adversarial-fix-plan. Does NOT re-investigate the diagnosis. Trigger as the review stage of a fix-pipeline, after adversarial-fix-plan has produced fix-plan.md.

fix-coder

Mechanical single-executor that applies a vetted fix plan to the codebase. Reads fix-plan.md (the Changes, Defenses, and Test Plan sections), translates file-level changes into concrete diffs, applies them, runs the test plan, validates that the bug is no longer reproducible (dev) or that the test exercising the failing path passes (prod), and produces a single fix commit. Stays strictly in scope: does NOT modify the plan, does NOT re-investigate, does NOT touch files outside the plan. Always writes fix-coder-report.md, even on failure. Trigger as the implementation stage of a fix-pipeline, after fix-plan-reviewer has approved fix-plan.md.

docs-refresh

Generates and maintains living documentation from the current codebase state. Reads a project-local manifest (.claude/docs-refresh.yml) that declares doc targets, sections, source inventories, and refresh boundaries. Uses marker blocks in docs to safely regenerate content without trampling manual prose.

command-center

The seat skill for Forge Command Center v2. Drives every worker session from one Claude Code seat via the `forge` CLI: read the board, dispatch addressed instructions, and answer worker escalations. Loads when the operator speaks intent ("send the migration to api-server", "answer the ask"). It drives the plumbing (bin/forge) — it never runs pipeline stages itself.

adversarial-verify

Single-agent verification skill that re-runs all tests and screenshots from a QA manifest, performs pixel-diff comparisons against baselines, and produces a structured verification report. This is the feedback loop that confirms fixes actually work and catches regressions. Trigger after adversarial-qa produced a manifest.yaml and the issues it found have been fixed, or as the verify step of an automated qa -> fix -> verify loop.

adversarial-qa

Adversarial QA framework using Claude Code Agent Teams. Two isolated QA testers independently test the running application from different angles with zero knowledge of each other, then a QA Synthesizer cross-verifies their findings, deduplicates issues, and produces a ranked issues report plus a test plan for missing coverage. Catches what unit tests miss - UI regressions, broken page loads, cross-stage interactions, API contract violations, integration failures. Trigger as the QA stage of a build pipeline, after implementation is committed.

开发工程师Skill与智能体开发

adversarial-proposal

Automated adversarial proposal framework using Claude Code Agent Teams. Spawns 3 persistent teammates (Proposer A, Proposer B, Synthesizer C) to investigate a technical problem through 4 rounds of isolated analysis, synthesis, critique, and reconciliation. A and B investigate the same problem independently with zero knowledge of each other; C reviews both and synthesizes; A and B critique C from their original contexts; C reconciles into a final plan. Trigger on: "adversarial proposal", "adversarial review", "multi-proposal", "debate approach", or explicit requests to investigate a bug/plan a feature/design an architecture with high confidence. For ambiguous phrases like "write a proposal" or "draft a plan", ASK whether the user wants the full adversarial flow or a single-pass hand-rolled document before proceeding. For small, bounded problems use `adversarial-lite` instead.

adversarial-lite

Lightweight adversarial proposal for low-risk, bounded tasks. 2-round version of adversarial-proposal: Sonnet investigators + Opus synthesizer. No feedback loop. Use for small features, clear bugs, focused refactoring. For high-risk, ambiguous, or multi-service problems, use adversarial-proposal instead. Trigger on: "lite proposal", "quick proposal", "adversarial lite", or when the lead confirms a problem is low-risk and bounded. Do NOT trigger on generic "investigate" or "propose" requests — those should use the full adversarial-proposal.

adversarial-investigate

Automated adversarial investigation framework using Claude Code Agent Teams. Spawns 3 persistent teammates (Investigator A, Investigator B, Synthesizer C) to diagnose the root cause of a known bug through 4 rounds of isolated analysis, synthesis, critique, and reconciliation. A traces FORWARD from the bug trigger; B traces BACKWARD from the failure point; both investigate the same evidence independently with zero knowledge of each other. C reviews both and synthesizes; A and B critique C from their original contexts; C reconciles into a final diagnosis. Allows "insufficient evidence" as a valid exit — does not force premature commitment to a cause. Trigger as the diagnosis stage of a fix-pipeline, after fix-reproducer (or directly off problem-statement.md if reproduce was skipped).

adversarial-implementation

Automated adversarial implementation framework using Claude Code Agent Teams. Takes a final-plan.md from the adversarial-proposal skill and produces a vetted implementation document with exact code diffs, test specifications, and a coverage matrix proving every plan item is addressed. Two agents work independently — a Surgical Implementer (minimal diffs) and a Coverage Guardian (complete tests) — then a Synthesizer merges them, both critique, and the Synthesizer reconciles into a final implementation.md. Use this skill after adversarial-proposal completes, or whenever the user wants to create an implementation document from a plan. Trigger on phrases like "create implementation", "implement the plan", "adversarial implementation", or "implementation doc".

adversarial-fix-qa

Automated adversarial QA framework for fix-pipelines using Claude Code Agent Teams. Spawns 3 persistent teammates (QA Confirmer A, Regression Hunter B, Synthesizer C) to verify a fix through 4 rounds of isolated testing, synthesis, critique, and reconciliation. A confirms the fix actually resolves the diagnosed bug under all conditions described in repro/diagnosis; B hunts for regressions in adjacent and downstream code affected by the fix; C synthesizes; A and B critique C from their original contexts; C reconciles into final fix-issues.md and fix-manifest.yaml. Trusts the upstream diagnosis, plan, and code — does NOT re-investigate or re-fix. Trigger as the QA stage of a fix-pipeline, after fix-coder has produced a committed fix.

adversarial-fix-plan

Automated adversarial fix-planning framework using Claude Code Agent Teams. Spawns 3 persistent teammates (Planner A, Planner B, Synthesizer C) to produce a vetted fix plan for a diagnosed bug through 4 rounds of isolated proposal, synthesis, critique, and reconciliation. A proposes a SURGICAL fix (minimum change to stop the diagnosed cause); B proposes a ROBUST fix (cause fixed + defenses against recurrence and adjacent fragility); C synthesizes; A and B critique C from their original contexts; C reconciles into a final fix-plan.md. Trusts the upstream diagnosis — does NOT re-investigate. Supports a revise mode for incorporating fix-review feedback. Trigger as the planning stage of a fix-pipeline, after adversarial-investigate has produced diagnosis.md.

proposal-reviewer

Independent review skill for critiquing a `final-plan.md` produced by the adversarial proposal framework. Performs a fresh pass over the problem statement, relevant code, and final plan to find incorrect root-cause assumptions, missing steps, hidden regressions, rollout gaps, and weak tests before implementation begins. Use this skill after `adversarial-proposal`, when the user wants an independent plan review, or when pointed at a proposals directory or `final-plan.md`.

forge-coder