mirror of
https://github.com/alirezarezvani/claude-skills.git
synced 2026-10-11 03:37:58 +00:00
46bb258a94
Deep audit of both domains against the AR v1 agentic-readiness rubric (audit/pm-product-agentic-2026-07/: master + per-domain reports + research-backed improvement fields + research digest), plus the harness layer the audit motivated: - pm-skills rebuilt as a context:fork orchestrator with an agentic delivery loop: pm_goal_router.py (8 lanes, exit-code route/ask/refuse), jira_snapshot_bridge.py (searchJiraIssuesUsingJql output -> Kanban Guide 2025 flow metrics with SLE + aging-WIP alerts + seeded Monte Carlo forecasts, or scrum-master sprint schema — verified end-to-end into velocity_analyzer.py), delivery_loop_gate.py (G1-G6 delegation governance: human owner, reviewer for agent tasks, machine-checkable acceptance, evidence-before-done, close refusal, exhausted-budget-is-escalation). - product-skills rebuilt as a context:fork orchestrator with the continuous-discovery loop: product_goal_router.py (16 lanes incl. standalone plugins), discovery_cadence_tracker.py (Torres weekly-habit health 0-100 with named gaps), ost_linter.py (O1-O5 Opportunity Solution Tree structural gates). - 6 new references citing 6-7 sources each (flow/forecasting canon, agentic delivery governance, PM loop playbook, continuous discovery, product operating model, AI product evals); pinned fixtures (expected_flow_metrics.json, sample OST/log). - cs-pm-orchestrator + cs-product-orchestrator agents; /cs:pm, /cs:grill-pm, /cs:pm-loop, /cs:product, /cs:grill-product, /cs:product-loop commands. - Fixed the two CLI-noncompliant product tools (user_story_generator.py, persona_generator.py): real argparse --help, seeded determinism, backward-compatible positionals. - Regenerated agent-harness manifests for both domains (orchestrators now score all five agentic_signals); updated domain CLAUDE.mds, plugin manifests (2.10.4), marketplace entries, and headline counters (602 tools / 731 references / 99 agents / 109 commands; derive_counters --check passes). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019Uzm8dKoeXPayJVMojpSbw |
||
|---|---|---|
| .. | ||
| .claude-plugin | ||
| agents | ||
| commands | ||
| skills/agent-harness | ||
| README.md | ||
agent-harness
Turn any domain folder of this repo into a bounded agentic loop: pick up a goal, compile it into tasks with machine-run verification, execute, verify, retry with caps, escalate to a human when budgets exhaust, and close only when everything is verified.
GOAL → goal_compiler → PLAN → loop_controller: [execute → verify]* → CLOSE
↑ retry ≤ caps, changed approach
└ ESCALATE — never fake success
What ships
| Piece | Purpose |
|---|---|
scripts/harness_manifest_builder.py |
Scan a domain folder → manifest.v1 JSON (skills, tools, checks, agentic signals) |
scripts/goal_compiler.py |
Goal + manifest → plan.v1 task plan; refuses vague goals (exit 3, forcing questions) |
scripts/loop_controller.py |
init/next/record/verify/close/status state machine; controller runs checks itself |
assets/harnesses/*.json |
18 committed per-domain manifests (regenerable, diff-stable) |
assets/harness_manifest.schema.json |
Manifest schema |
references/ |
Agentic-loop canon, verification discipline, domain-harness design (cited) |
agents/harness-runner.md |
Stateless one-task-per-invocation executor |
commands/cs-harness.md |
/cs:harness <domain> <goal> end-to-end driver |
All tools are stdlib-only, pass --help and --sample, and emit JSON.
Design lineage
Anthropic's long-running-agents harness (feature-list + stateless shifts), verifier's law,
SWE-agent's environment-feedback lesson, Ralph-loop fresh-context iteration, Cognition's
serialize-writers rule, and this repo's own tc-tracker / autoresearch locked-evaluator /
loop-library stop-state primitives. See skills/agent-harness/references/.