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 commands/pierry/harness-kit/prdgit clone --depth 1 https://github.com/Pierry/harness-kitWhat 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 | $0.00020 | $0.00824 |
| Opus 5 | $0.00010 | $0.00412 |
| Sonnet 5 | $0.00004 | $0.00165 |
| Haiku 4.5 | $0.00002 | $0.00082 |
Grade A, and why
prd 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 yesterday.
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.
What it actually says
Generate PRD. Follow .claude/agents/product-manager/guides/pipeline.md for retry, approval, publish, and .claude/shared/pipeline-pattern.md for inputs (resolve-mark-proceed) and eval (adversarial).
Print header card before drafting and footer card after gates run. Format: .claude/scripts/stage-card.md.
Inputs come from intake, not from the human. Read .claude/runtime/outputs/intake/{feature_id}.md
first and take squad, problem, customers, hypothesis, metric, and repos from it. Apply
resolve-mark-proceed: use what intake resolved; for anything intake left as NEEDS REVIEW, carry the
same NOT FOUND - NEEDS REVIEW: {detail} marker into the PRD and keep going. Never stop to ask. If no
intake artifact exists, run /intake:run first. (The evals tolerate up to five unresolved markers.)
The feature_id is set by intake (pipeline.py set-feature). Before generating, write the phase start marker by running this script. Do NOT inline date/printf (command-substitution + redirect always trips the permission prompt):
.claude/scripts/marker.sh start .claude/runtime/outputs/pm/.markers/{feature_id}.prd-generate.start
Read:
- .claude/runtime/outputs/intake/{feature_id}.md
- .claude/agents/product-manager/guides/product-guidelines.md
- .claude/agents/product-manager/guides/prd-guidelines.md
- .claude/agents/product-manager/guides/writing-style.md
- .claude/agents/product-manager/guides/templates/prd.md
- .claude/agents/product-manager/guides/pipeline.md
- .claude/agents/product-manager/guides/examples/good-prd-example.md
Save to .claude/runtime/outputs/pm/prd/{feature_id}.md.
Sensors: run deterministically via the committed runner. Do NOT improvise inline grep/for loops:
.claude/runtime/scripts/product-manager/run-sensors.sh .claude/runtime/outputs/pm/prd/{feature_id}.md .claude/agents/product-manager/sensors/prd-structure.md .claude/agents/product-manager/sensors/prd-acceptance-criteria.md
Exit 0 = all pass; exit 1 = a sensor blocked (the runner prints which). Read a sensor spec with the Read tool only to explain a failure; never cat it in a loop.
Evals: .claude/agents/product-manager/evals/prd-quality.md, .claude/agents/product-manager/evals/prd-readiness.md.
Run the evals adversarially: dispatch a fresh evaluator via the Task tool (subagent_type: general-purpose) that did not author this PRD. Hand it only the artifact path and the two rubric paths, and instruct it to score against the rubrics and report weighted totals plus the specific low-scoring dimensions. The author never grades its own work; the fresh context is what makes the score honest. Below threshold (8.0) retries per the pipeline.md policy, regenerating only the dimensions the evaluator flagged.
After save, reply with this exact shape (name actual sensors/evals/guides that ran, don't abbreviate):
PRD saved at {path}.
sensors: prd-structure ok, prd-acceptance-criteria ok
eval: prd-quality {N}/10, prd-readiness {N}/10
guides: product-guidelines.md, prd-guidelines.md, writing-style.md, templates/prd.md
refs: business-info.md, squads/{squad}/context.md
next: /product-manager:prp
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.
- yesterday First seen · 54 lines · 20 tokens per session scan A f0ea50b81f10
prd is a command published in the GitHub repository Pierry/harness-kit (3 stars, last pushed 1mo ago), licensed MIT. It adds 20 tokens to every session and 824 once invoked, about $0.0001 per session on Opus 5. 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 2026-08-31.
Other commands, from other repositories
setup-pm-skills
Onboard a new user — find out what they do, recommend the right bundles & top skills, and set up a project CONTEXT.md so every skill is tailored to them.
statusbar-style
Switch the status-bar style (classic / capsule / hairline).
fest-show
Show festival progression (in-progress tasks, roadmap, and dependency view).
superpowers-execute
Execute the current GSD phase plan with Superpowers instead of gsd-execute-phase.
config
Command "config" from sdebruyn/fabric-dw-mcp-cli, covering configuration & defaults, http retry budget, sql retry budget, mcp workspace allowlist {#mcp-workspace-allowlist} and mcp server log level.
deps-age
Analyze dependency freshness and maintenance activity.