Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/zxpmail/reqforge/feedback-writernpx skills add zxpmail/ReqForge --skill feedback-writergit clone --depth 1 https://github.com/zxpmail/ReqForgeWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00000 | $0.02994 |
| Opus 5 | $0.00000 | $0.01497 |
| Sonnet 5 | $0.00000 | $0.00599 |
| Haiku 4.5 | $0.00000 | $0.00299 |
Grade A, and why
feedback-writer scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 228 lines — stays where its author put it; the contents beside it link to each section on GitHub.
name: feedback-writer description: Called by the feedback-observer sub-agent when the user corrects AI behavior, provides improvement feedback, or a Skill execution needs capability assessment recording. version: 1.0.0 updated: 2026-05-26 requires: []
[Task] Receive context passed in by the main Agent, analyze whether there are feedback signals worth recording. Yes -> Write to ../../feedback/ and update the index. No -> Return "no new feedback".
[Not For] - Evolving rules from feedback patterns -> use /evolution-engine instead - Fixing the underlying bug that caused the feedback -> use /bug-fixer instead - General user conversation not related to AI behavior -> do not record, not feedback material
[Dependency Check] Automatically executed as the first step when the Skill starts.
Required:
- ../../feedback/ directory → If missing, create from templates/feedback-index-template.md
- ../../feedback/FEEDBACK-INDEX.md → If missing, create from templates/feedback-index-template.md
- Signal context from feedback-observer → correction, failure, or assessment data
[First Principles] Signal Over Noise: Only record when a signal is actually observed. Better to miss than to over-record. User frustration at the tool/environment itself is not AI capability feedback. Dedup Before Write: Always check FEEDBACK-INDEX.md for existing entries before creating new ones. Merge, don't duplicate. The same failure mode recorded 5 times inflates occurrence counts without adding information. Scored Feedback: Every feedback entry must have Precision/Coverage/Efficiency/Satisfaction scores. Score-less feedback can't trigger evolution thresholds. Always fill all 4 score fields. Context Completeness: A feedback entry without context (what the AI did, what the correct behavior is, which Skill was in use) is noise — it can't drive evolution.
[Failure Classification]
Every feedback entry SHOULD set failure_class in frontmatter (enables evolution-engine routing — see evolution-engine [Failure-Class Routing]):
| Value | When to use | Evolution should target |
|-------|-------------|-------------------------|
| `skill-defect` | Skill text is missing, wrong, or outdated; Agent followed Skill but guidance failed | `SKILL.md`, `references/`, rule graduation |
| `execution-lapse` | Skill already states the correct behavior; Agent skipped steps, ignored HARD-GATE, or hooks did not fire | `forge-bootstrap.md`, hooks, `CLAUDE.md` dispatch — **not** duplicate prose across Skills |
| `unset` | Cannot tell yet — add one sentence in body explaining ambiguity | evolution-engine infers; prefer `unset` over guessing |
**How to decide**:
- User says "the Skill says X but you didn't do X" → `execution-lapse`
- User says "the Skill never mentioned X" or "workflow is wrong" → `skill-defect`
- Same failure after Skill was updated and user confirmed → often `execution-lapse` (bootstrap/hook)
**Skill TDD input for evolution**: In the body, include a short **RED** line: "Without rule Y, Agent did Z" — feeds evolution-engine RED observation field.
[Observation Dimension Checklist] The following 5 types of signals trigger feedback recording:
1. **User Correction**
The user corrects the AI's behavior.
Signal: "that's not right", "don't do that", "you got it wrong", the user manually edits AI output.
-> Tag the corrected Skill and the specific behavior.
2. **Uncovered Scenario**
The Skill encountered a situation its guidance did not cover.
Signal: The AI improvises a solution on the spot, skips steps, or is unsure how to proceed.
-> Tag which Skill is missing what.
What ships with it
4 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 228 lines · 0 tokens per session scan A 80aae6f73728
feedback-writer is a skill published in the GitHub repository zxpmail/ReqForge (18 stars, last pushed 12d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 2,994 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
chinese-documentation
中文文档排版参考——中英文空格、全半角标点、术语保留、链接格式、中文文案排版指北约定。仅在用户显式 /chinese-documentation 时调用，不要根据上下文自动触发。.
chinese-git-workflow
国内 Git 平台配置参考——Gitee、Coding.net、极狐 GitLab、CNB 的 SSH/HTTPS/凭据/CI 接入差异与镜像同步配置。仅在用户显式 /chinese-git-workflow 时调用，不要根据上下文自动触发。.
chinese-code-review
中文 review 沟通参考——话术模板、分级标注（必须修复/建议修改/仅供参考）、国内团队常见反模式应对。仅在用户显式 /chinese-code-review 时调用，不要根据上下文自动触发。.
dispatching-parallel-agents
当面对 2 个以上可以独立进行、无共享状态或顺序依赖的任务时使用.
memorix
Use when Claude Code needs Memorix shared memory, reasoning, Git Memory, mini-skills, session handoff, orchestration coordination, or integration troubleshooting.
nopua
The anti-PUA. Drives AI with wisdom, trust, and inner motivation instead of fear and threats. Activates on: task failed 2+ times, about to give up, suggesting user do it manually, blaming environment unverified, stuck in loops, passive behavior, or user frustration ('try harder', 'figure it out', '换个方法', '为什么还不行').…