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 skills add lukasrepublic/agentic-foundry --skill research-capturegit clone --depth 1 https://github.com/lukasrepublic/agentic-foundryWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/lukasrepublic/agentic-foundry/research-capture)<a href="https://agentmods.dev/skills/lukasrepublic/agentic-foundry/research-capture"><img src="https://agentmods.dev/badge/skills/lukasrepublic/agentic-foundry/research-capture/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/lukasrepublic/agentic-foundry/research-capture"><img src="https://agentmods.dev/badge/skills/lukasrepublic/agentic-foundry/research-capture.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What 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 5.1 | $0.00147 | $0.01453 |
| Opus 5.5 | $0.00059 | $0.00581 |
| Sonnet 5.5 | $0.00029 | $0.00291 |
| Haiku 4.5 | $0.00015 | $0.00145 |
Grade A, and why
research-capture 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 1mo 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 — 77 lines — stays where its author put it; the contents beside it link to each section on GitHub.
/foundry:research-capture — durably persist the research gate's full evidence trail
The research-first record phase writes a distilled decision record (at the configured
governance.decisions_path, defaults to .foundry/decisions/) — the why. But the full claim-level evidence behind a deep-research run (every verified claim, its
adversarial vote, the source table, and crucially the refuted / killed claims) lives only in an
ephemeral workflow-output file and is lost on cleanup. So a later auditor asking "was this 'industry
consensus' actually real?" has only the author's summary. This discipline closes that gap the standard
way — a durable template + an author convention, exactly as the ADR / decision-record ecosystem
does (adr-tools / log4brains / MADR ship new/link scaffolders + templates; none ship a
byte-identical emitter or a fail-closed completeness classifier).
It is a DISCIPLINE, not a code helper (the deep-spec-audit re-ground)
The v1 design — a deterministic byte-identical capture helper + a fail-closed completeness validator —
was novel machinery for a solved problem (a deep-spec-audit category-error). Determinism / byte-identity has no
prior-art basis for a human-read prose doc (it governs machine-consumed attestations, not narratives).
The hand-written .foundry/research/wf_c3c6df43-3b5-machinery-view.md — complete, with a Refuted/Killed
section + both-way ADR links, authored with zero new code — is the decisive disproof. So capture is
authoring from a template; the lightweight drop-in wiring lint this line originally shipped
was removed with the drop-in-check registry and was not ported (see the note below) — capture
ships no code.
The discipline
At a deep-research-bearing research-first gate (a multi-source run, not a mechanical/obvious-standard decision):
- Persist the full trail. Author
<run-id>-<slug>.mdat the configured governance-record research location —governance.research_pathin.claude/foundry-project.json(defaults to.foundry/research/) — fromresearch-artifact-template.md(this skill's sibling). The<run-id>-<slug>filename is path-safe ([a-z0-9._-]). Required sections: the run-id, the question, the stats (sources / claims / verified / confirmed / refuted), the verified claims, the## Refuted / disconfirming evidencesection (REQUIRED — present even if(none)), the sources table, and a both-way link with the decision record. - Retain the disconfirming evidence in the ADR too. Author the ADR — at the configured
governance-record decisions location (
governance.decisions_pathin.claude/foundry-project.json, defaults to.foundry/decisions/) — with a## Refuted / disconfirming evidencesection (Nygard Consequences-style) so the killed/disconfirming claims are retained in the decision record itself as well as the linked artifact. - Link both ways. The ADR points at the artifact; the artifact points at the ADR.
What ships with it
1 file 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.
- 1mo ago First seen · 77 lines · 147 tokens per session scan A 33abe5602dfb
research-capture is a skill published in the GitHub repository lukasrepublic/agentic-foundry (1 stars, last pushed yesterday), licensed MIT. It adds 147 tokens to every session and 1,453 once invoked, about $0.0006 per session on Opus 5.5. 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 2026-09-03.
Other skills, from other repositories
atlas-report
Render agent findings as a styled HTML report in the browser. Use when asked for "full report", "detailed report", "show in browser", or when CLI output exceeds the 40-line budget.
g-docs
Documentation audit and generation. Scans for missing or stale code docs, missing README sections, undocumented env vars, CHANGELOG gaps, and architectural decisions without ADRs. Targeted scope fixes gaps immediately via doc-writer. Whole-codebase scope produces a prioritised documentation debt report and optional…
document-pdf
Handles PDF/UA accessibility, real redaction, metadata scrubbing, signing, and extraction. Use when redacting or signing PDFs; for plain create/merge prefer the official pdf skill.
document-xlsx
Builds Excel charts, pivots, validation, exports, Office Scripts/Graph automation, and workbook audits. Use when automating workbooks; for plain edits prefer official xlsx skill.
document-docx
Builds docxtpl templates, DOCX extraction for RAG, and EAA accessibility checks. Use when templating Word or extracting tables; plain edits and redlines: official docx skill.
volt-firmware
Produce a complete firmware architecture spec for a described device — layer diagram, module responsibilities, HAL interface definitions, key state machines, RTOS decision. Use when asked to "design firmware architecture", "plan embedded firmware", "architect an IoT device", "how should I structure this firmware", or…