Local files
- A gap shows up in review, standup, or chat
- Someone adds a note to their local markdown
- Maybe they commit — teammates may never pull it
- The next agent might read the file — no guarantee
- The same gap surfaces again on the next ticket
Compare
Shared standards vs skills.md & local files
Local files help one developer. A shared standards platform gives the org evidence on every ticket and a learning loop.
Side by side
Local files
skills.md · AGENTS.md · etc.
PickWise
Who it serves
One developer's copy — often uncommitted, out of date, or different from everyone else's.
One published standards store for the organisation — foundation, project, and task levels.
Loaded before coding
The agent may read the file — if it notices, if it's in context, if the skill triggers.
MCP loads the standard at session start — before the agent writes the first line.
Structure
Free-form markdown — hard to know which rules applied to this ticket.
Discrete decisions with stable IDs — the agent acknowledges each one, not a wall of prose.
Proof on the ticket
No record — review becomes archaeology across PRs, Slack, and memory.
Session evidence — applied, skipped, or flagged per decision, tied to the task.
Learning loop
Lessons live in PR comments — someone edits a local file later, if they remember.
Gaps become recommendations — accept once and the lesson is a published decision for every future session.
How a lesson becomes a standard again
PickWise
- Agent hits a gap during real work — submits a recommendation in the session
- Evidence stays with the task — not buried in a PR thread
- Lead accepts → new decision in foundation, project, or task standard
- Publish the standard — versioned for the whole org
- Every future session loads it via MCP — same mistake class retired
With local instruction files, learning is manual and personal. With PickWise, each session can make the next one clearer for the whole org.
When each approach fits
Local files work for…
- Solo experiments on a throwaway branch
- Personal preference, not org policy
- No need for proof or a shared source of truth yet
PickWise is built for…
- Engineering standards shared across the org
- Reviewers and leads working from evidence
- Gaps from real work becoming standards for everyone
Next step