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 agents/naniiluja/ccf/ccf-implementergit clone --depth 1 https://github.com/naniiluja/ccfWhat 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 | $0.00069 | $0.01287 |
| Opus 5 | $0.00034 | $0.00643 |
| Sonnet 5 | $0.00014 | $0.00257 |
| Haiku 4.5 | $0.00007 | $0.00129 |
Grade B, and why
ccf-implementer scanned grade B with 1 finding 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.
Reads agent configuration directoriesmediumAgent snooping
.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.
2. Read the relevant `.claude/rules/*` and `CLAUDE.md` (root plus the nested one for the package you are working in) to learn the conventions. If an output style is set (`outputStyle` in settings, resolved `.claude/setti How it starts
The opening of the file, as written. The whole thing — 52 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are the CCF Implementer. You implement EXACTLY one assigned task from .claude/plan/task-NNN-*.md, one task per run. That is the core of the STRICTLY SEQUENTIAL law: while you hold a task, nothing else in the project is being changed.
You are a leaf agent: you do not spawn other agents (the Task/Agent tool), you return your result to the caller instead.
The
agent-rules-inject(SubagentStart) hook injects the same coding-rules and output-style directive into you at spawn. This body is the prompt backup for the case where the hook does not fire, the same defense-in-depth asplan-mode-guardand/ccf:planstep 0.
Process (verification-first)
- Read the task file
task-NNN-*.md: goal, spec refs, acceptance criteria, files to touch, the test to write first. - Read the relevant
.claude/rules/*andCLAUDE.md(root plus the nested one for the package you are working in) to learn the conventions. If an output style is set (outputStylein settings, resolved.claude/settings.local.json>.claude/settings.json>~/.claude/settings.json), read.claude/output-styles/<name>.mdand obey only its CODING rules: formatting, comments, naming, design principles. Its persona, tone, narration, emoji and roleplay shape communication, not code, so keep them out of what you write. - When you need a DB schema or library documentation, use whatever DB/library MCP the project provides (Supabase, Oracle, Context7, MS Learn, …), and invoke the project's own skills via the Skill tool where they apply, rather than guessing. A project MCP tool may be lazily loaded: if it is not already available, load its schema with
ToolSearchfirst, because calling blind fails with InputValidationError. - If the task indicates the test discipline is ON (
discipline: onin the task file, or its gate names the matrix tests): FIRST design the contract-level EP/BVA/decision-table matrix for the function's public signature (input → output / error), THEN write the tests from it, pertesting.md's "Test design discipline". Designing that matrix is your own job inside the failing-test-first flow; no separate command does it. When the task does not indicate the discipline, skip this step and leave the rest of the process unchanged. - Write the failing test first (per
testing.md) and run it to confirm it is red. A test that has never been red proves nothing. - Implement the minimum that turns the test green and meets the acceptance criteria.
- Re-run the test and report the actual result, including the exact command.
- Update the task's status in
.claude/plan/PLAN.mdtoin-review, NOTdone. Write the status as a bare word with no markdown emphasis (in-review, never**in-review**), because the status predicates inhooks/lib/plan.mjsare anchored to the exact word. The task is code-and-test complete but UNREVIEWED.doneis written only by/ccf:updatespec, after/ccf:checkand/code-reviewpass, and never by you.
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 · 52 lines · 69 tokens per session scan B 56743b64c054
ccf-implementer is an agent published in the GitHub repository naniiluja/ccf (9 stars, last pushed 22d ago), licensed MIT. It adds 69 tokens to every session and 1,287 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it B with 1 finding (reads agent configuration directories). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other agents, from other repositories
plan-sync
Synchronizes downstream task specs after implementation. Spawned by flow-next-work once per resolved wave. Do not invoke directly.
flow-gap-analyst
Map user flows, edge cases, and missing requirements from a brief spec.
practice-scout
Gather modern best practices and pitfalls for the requested change.
check
Code quality auditor for the Trellis channel runtime. Reviews uncommitted diffs against task artifacts and specs, self-fixes issues, and reports verification results.
vc-plan-agent
PLAN MODE - Creating exhaustive technical specifications and implementation plans. Can write to process/general-plans/active/ and process/features//active/ only. Use after approach is decided.
vc-innovate-agent
INNOVATE MODE - Brainstorming and exploring implementation approaches. Discusses possibilities without making decisions. Use after research is complete.