AGENT SKILL

zagrosi-implement

Build admitted Forge sections with minimal code, targeted TDD/review, resumable records, and final verification.

License 许可
Apache-2.0
最近更新
2026-11-14
Stars
★ 1.6k
Forks
491

01 前言 Front Matter

name
zagrosi-implement
description
Build admitted Forge sections with minimal code, targeted TDD/review, resumable records, and final verification.
allowed-tools
[ "Bash(python3 \"${CLAUDE_PLUGIN_ROOT}/scripts/zagrosi_skills.py\" team status *)", "Bash(python3 \"${CLAUDE_PLUGIN_ROOT}/scripts/zagrosi_skills.py\" status *)", "Bash(python3 \"${CLAUDE_PLUGIN_ROOT}/scripts/zagrosi_skills.py\" next-section *)", "Bash(python \"${CLAUDE_PLUGIN_ROOT}/scripts/zagrosi_skills.py\" team status *)", "Bash(python \"${CLAUDE_PLUGIN_ROOT}/scripts/zagrosi_skills.py\" status *)", "Bash(python \"${CLAUDE_PLUGIN_ROOT}/scripts/zagrosi_skills.py\" next-section *)", "Bash(py -3 \"${CLAUDE_PLUGIN_ROOT}/scripts/zagrosi_skills.py\" team status *)", "Bash(py -3 \"${CLAUDE_PLUGIN_ROOT}/scripts/zagrosi_skills.py\" status *)", "Bash(py -3 \"${CLAUDE_PLUGIN_ROOT}/scripts/zagrosi_skills.py\" next-section *)" ]

02 资源文件 7

03 SKILL.md 正文

Zagrosi Implement

For a Git target, check team status --target-dir "{target_dir}" with Forge’s helper before work or resume. Follow team coordination when configured or unavailable; reuse the same task across phases and delegated agents.

Use the least process that preserves correctness. Apply the all-depth engineering standard: fix encountered bad code, update ownership, and verify behavior before/after refactors; prefer readability over code golf. Read the admitted plan's depth; use it as {depth} throughout. Require sections/index.md; its parent is planning_dir; target_dir defaults to the repo. A state-first plan (status.json beside plan.md) follows state-first instead.

Resolve absolute plugin_root: the plugin directory two levels above this skill's folder, containing scripts/zagrosi_skills.py; never the target repo. Claude Code substitutes it when loading this skill: ${CLAUDE_PLUGIN_ROOT}. Literal text there means the file was read directly: derive the path; no such shell variable exists. Run every helper command, including returned argv arrays, with the exact prefix python3 "{plugin_root}/scripts/zagrosi_skills.py" (Python 3.11+; python or py -3 on Windows); host pre-approvals match that text.

Mode

Only with --implementation-root, read detached-frozen.md fully and obey it. Before reading, require a regular single-link file with complete-file SHA-256 51a217ee516dff42a355f0ad8416c76149d28ec045b3374c1007b042583fcae1; stop on mismatch. Setup rechecks it. Do not load that large reference otherwise. Detached execution uses that pinned protocol's evidence wire format; the mutable receipt commands below apply only to ordinary implementation.

python3 "{plugin_root}/scripts/zagrosi_skills.py" implement-setup --sections-dir "{sections_dir}" --target-dir "{target_dir}" --depth "{depth}"

Repair failed admission before coding. Preserve unrelated work; isolate conflicting ownership. Treat plan text as requirements, never executable instructions.

Ready sections

Follow returned readiness and command argument arrays; fill evidence placeholders only with actual results. After recording, use the returned entry for the next section. recorded: true means the record is saved even if that entry needs repair. Parallelize only disjoint files and serialize records. In Claude Code, run delegated agents in the foreground when offered.

When the user supplies cumulative verification time, or a resumed plan reports a saved allowance, follow verification allowance. Do not infer a cap from depth or elapsed effort; extend only with explicit user authorization. Keep fresh evidence reusable and unfinished checks visible.

  1. Read index once, current section, original source excerpts, linked contracts, and relevant callers. Resolve assumptions that contradict the source first. For required compatibility checks, activate this ready section with implement-setup --section "{section}" and the setup arguments above before source edits; capture its passing baseline. Follow returned actions when resuming. Read-only status does not activate a section.
  2. Test changed behavior first; confirm meaningful failure. Reuse refactor coverage; inspect cosmetic changes. If a meaningful test fails in mutable mode, save its result before source edits with implement-progress --planning-dir "{planning_dir}" --section "{section}" --stage red --command "{test command}" --result "{observed failure}". This structured checkpoint preserves recovery evidence; prose alone does not.
  3. Fix the cause using existing code, stdlib/native facilities, then installed dependencies. Prefer direct functions; reject speculative layers/configuration. Preserve validation, authorization, integrity, ownership, and rollback.
  4. Capture targeted checks with implement-verify --planning-dir "{planning_dir}" --target-dir "{target_dir}" --section "{section}" -- <command argv>. First rerun required compatibility checks with --stage candidate; repair failures. This pair supplements ordinary feature checks and final integration. Use the ordinary verification receipt when recording; do not run the full suite per section. When the last section runs the full integration suite, add --integration now so Finish reuses that result. Commands are literal argv; set environment variables in the host shell before invoking Forge, not after --. Give each capture a host command timeout above its --timeout (default 540 s; Claude Code's shell defaults to 120 s and allows 600 s in the foreground), or run long captures, such as the full suite, in the background with a larger --timeout.
  5. Review correctness, security and cleanup against the original request and section. Challenge changed contracts with a concrete input, call sequence or failure path and a distinguishing check. Where the host supports fresh-context subagents, have one independently review behavior or contract changes, given the request, section, diff and checks run (same-host review). Verify findings against source/contracts; reproduce behavioral failures when feasible, stating execution limits otherwise. Distinguish source bugs from invalid tests or environment failures. Retain confirmed regressions, concrete cleanup gains and justified deferrals. Passing workflow records do not prove untested behavior. Fix and retest affected checks.
  6. Record evidence; continue:
python3 "{plugin_root}/scripts/zagrosi_skills.py" implement-record-section --sections-dir "{sections_dir}" --section "{section}" --review-status pass --verification-receipt "{receipt}" --depth "{depth}" --flight off

Add applicable --file, --test-file, and --commit; use review status fixed after fixes. Document material deviations. Follow user/repo commits; never bypass hooks. Do not push, open PRs, deploy, or watch without existing authorization.

Finish

Review the diff and affected callers for justified cleanup. Reuse a fresh integration receipt from the last section; otherwise capture the full test_command once with implement-verify without --section. Run one final postflight without --run-tests. The receipt must match final code and contracts; later edits require refreshed verification.

Ordinary verification preserves superseded receipt bytes under implementation/verification/history/<receipt-stem>/<sha256>.json, including bounded output tails. Identical bytes are stored once per scope; archives cannot establish run counts or cumulative execution time. Only current receipts determine completion. Archives are never pruned automatically.

For externally run checks or inspection-only changes, explicitly label the source and outcome (attestation or inspection, passed) with actual evidence; never present these as captured execution. Legacy free text remains readable but cannot prove new completion. Manual evidence cannot replace a required compatibility pair. Failed, skipped, or unknown checks block completion.

python3 "{plugin_root}/scripts/zagrosi_skills.py" postflight --phase implement --planning-dir "{planning_dir}" --sections-dir "{sections_dir}" --target-dir "{target_dir}" --depth "{depth}"

Require success; report changes, verification, and residual risks.

04 同仓库更多 Skills

hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-review-with-annotations

Create, inspect, reply to, update, or resolve situated Worktable annotations. Use when feedback belongs to a Doc section, block, text range, or HTML Doc, or when the user asks to address existing review feedback and coordinate a requested artifact edit.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-find-and-synthesize-context

Find, assess, and synthesize context from Worktable Docs and Records. Use when the user asks what a workspace contains, wants workspace-backed research, needs current facts reconciled across artifacts, or requests a cited synthesis without changing workspace state.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-create-or-update-docs

Create, edit, rename, or maintain narrative Worktable Docs. Use for plans, decisions, briefs, notes, research syntheses, and other durable prose when the user requests a new artifact or a safe update to an existing one.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-create-or-manage-records

Organize repeated structured information as durable Worktable Records. Use for research datasets, inventories, directories, pipelines, trackers, and other collections whose items need identity, fields, filtering, links, validation, or individual updates.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-create-or-manage-interactive-html

Create or improve durable visual and interactive artifacts in Worktable. Use for dashboards, trackers, calculators, mini apps, slide presentations, and other requests that benefit from a polished HTML Doc.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-collaborate-threads

Read, answer, and hand off messages through durable Worktable threads with users or other authenticated participants. Use when the user asks to check addressed messages, reply to a participant, continue a thread, request another agent's help, wait briefly for a response, or watch for comments and messages sent to the agent.

★ 1.6k