AI Skills Library v1

10 reusable prompts that compound across every agent

ironbark sits in your tenant and harvests every Claude Code session. When the same prompt pattern shows up twice, it lifts it into the library — automatically, without anyone curating. Other agents in your fleet pick it up on next deploy.

Skills harvested

Total agent runs

1,139

reusing these prompts

Pharmacy

2 skills

Drug Name Normaliser

Harvested 2026-04-12 · 47 reuses

Maps brand names to generic INNs and back. Handles common typos and trailing whitespace.

Use when: Any pharmacy agent that compares medication_master rows against external prescriptions or vendor catalogues.

Show prompt snippet

Given a medication name (possibly a brand, possibly a generic), return: { generic_name, brand_names[], schedule, dosage_form }. If ambiguous, return the top 3 candidates with confidence scores. Trim whitespace; strip strength suffixes for the lookup.

Used by:pharmacy-reorder prescription-dispense

Supplier Resolver

Harvested 2026-05-08 · 12 reuses

Maps a medication to its preferred + fallback suppliers. Honours credit-day terms.

Use when: Pharmacy reorder; any procurement agent that drafts POs.

Show prompt snippet

Given a medicine_id and quantity, return the supplier order: [primary, fallback_1, fallback_2]. Use medicine_master.supplier_id as primary. Fallbacks: same-category suppliers ranked by credit_day descending. Flag if credit_limit will be exceeded.

Used by:pharmacy-reorder

Engineering

2 skills

Markdown Frontmatter Validator

Harvested 2026-04-15 · 312 reuses

Reads a Markdown file, validates the YAML frontmatter against the schema, returns errors with line numbers.

Use when: GitHub Actions sync workflow; pre-commit hook in the vault.

Show prompt snippet

Validate the YAML frontmatter against this schema: { department: enum, agent_pack?: string, sensitivity: enum, last_reviewed: date, owner: email, status: enum, tags: array }. Return { valid: bool, errors: [{ field, message, line }] }.

Used by:sync-vault.yml pre-commit-hook

Scoring Checklist Generator

Harvested 2026-04-18 · 28 reuses

Given an agent's stated goal and 2-3 example outputs, generates a 5-7 item scoring checklist.

Use when: Step 7 — when defining 'good enough' for a new agent's outputs.

Show prompt snippet

Read the agent's goal + 3 example outputs. Generate 5-7 specific, falsifiable checks every output must pass. Each check must be answerable yes/no without judgement. Format: must_pass: [list], score_threshold: 'X/Y', flag_for_review_if: [conditions].

Used by:pharmacy-reorder content-brief-to-draft lead-research weekly-pnl

Marketing

1 skill

Brand Tone Matcher

Harvested 2026-04-20 · 89 reuses

Reads your brand voice doc + 3 sample outputs and rewrites a draft to match. Stays under specified word count.

Use when: Content brief → draft, social media, customer-facing emails.

Show prompt snippet

Read [brand_voice.md] and these 3 reference samples. Rewrite the input draft to match the voice. Constraints: max [N] words, grade-8 reading level, no jargon undefined in [glossary.md].

Used by:content-brief-to-draft support-triage

Sales

2 skills

ICP Fit Scorer

Harvested 2026-04-22 · 134 reuses

Scores a prospect 0-100 against your ICP. Justifies the score with cited evidence.

Use when: Pre-meeting research. Lead qualification. Territory planning.

Show prompt snippet

Read [icp_definition.md]. Given the prospect's company website + LinkedIn + recent news (provided), score 0-100. Output: { score, top_3_signals: [], top_3_anti_signals: [], suggested_questions: [] }. Cite sources.

Used by:lead-research proposal-drafter

Expense Categoriser

Harvested 2026-05-02 · 56 reuses

Classifies a receipt into a chart-of-accounts category with > 80% confidence.

Use when: Bookkeeping automation; expense reports; budget tracking.

Show prompt snippet

Given the OCR'd receipt + vendor name + amount, classify into [chart_of_accounts.md]. Confidence > 80% to ship; below = flag for human. Use vendor history (last 30 entries provided) to disambiguate.

Used by:expense-categorisation

Compliance

1 skill

PHI Redactor

Harvested 2026-04-30 · 203 reuses

Removes patient-identifying fields from any text/JSON before it leaves the perimeter for an external LLM.

Use when: Mandatory wrapper around any agent that touches patient data — runs BEFORE the LLM call.

Show prompt snippet

Detect and replace: patient names → [PATIENT_N], medical record numbers → [MRN_N], DOBs → [DOB_N], addresses → [ADDR_N]. Maintain a per-call mapping for re-hydration on the response. Return { redacted_text, mapping }.

Used by:clinical-summary-agent discharge-letter-drafter

Communications

2 skills

Meeting Action Extractor

Harvested 2026-04-25 · 167 reuses

From a call transcript, extracts action items, assigns owners by name+role, sets due dates if mentioned.

Use when: Meeting summary agent; client onboarding kickoff calls.

Show prompt snippet

Read transcript. Extract action items in the form: { task, owner_name, owner_role_guess, due_date_or_null, source_quote }. Skip 'someone should' / 'we could' wording — only commitments.

Used by:meeting-summary client-onboarding

Email Thread Summariser

Harvested 2026-04-28 · 91 reuses

Condenses a long email thread into a 3-bullet status + 1 next-step recommendation.

Use when: Inbox triage; client onboarding handoffs; pre-meeting prep.

Show prompt snippet

Summarise this email thread into: 3 bullets (current status + 2 most recent decisions), 1 recommended next action with owner. Cite the message that drove each bullet (date + sender).

Used by:support-triage client-onboarding

Run ironbark in your tenant

Open source. Sits in your repo. Harvests skills from every Claude Code session and syncs back to the community library every 30 min.

GitHub repo Step 4 sets it up