Write

Context engineering framework for AI coding agents — Write, Isolate, Select, Compress

Install

npx -y skills add ph3on1x/wisci --skill write

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Externalizes knowledge from the conversation into persistent, staleness-tracked markdown files under .wisci/context/. Use when research results, decisions, or findings are worth keeping beyond this session, or when the user wants context saved to disk.

SKILL.md

3.4 KB, as published. Nobody here has run it

/write — Context Externalizer

Move knowledge from the context window into a structured, persistent file in .wisci/context/. This is the "save to disk" operation — preserve what would otherwise be lost to context limits or session endings.

Boundary: /write stores reusable knowledge (research, decisions, architecture notes). Session work state (goal, in-progress, next steps) belongs in /compress.

Timestamp: !date '+%Y-%m-%d %H:%M'

Execution Flow

Parse arguments. Read $ARGUMENTS to determine what to externalize from the current context.
Derive topic slug. Extract the core topic noun phrase (strip filler: "results on", "notes about"), kebab-case, lowercase, max 5 words. If the user names a target file explicitly, use it.
Check for existing file. Glob .wisci/context/ for the slug:
- No match: Create Mode below.
- Exact match, same topic: Merge Mode — read and follow references/merge-procedure.md.
- Exact match, different topic: ask the user — merge or disambiguated new name.
Write to .wisci/context/<topic-slug>.md (create directories as needed).
Confirm. Report path, created vs merged, and a one-line content summary.

Create Mode

Extract and structure the information from the current context per $ARGUMENTS
Organize into logical ## sections
Build a ## References section listing every file path mentioned in the content
Write using the output format below

Output Format

# <Title inferred from arguments>

> Last updated by /write on <timestamp from above>
> Source: <what session context this was extracted from>

## Summary
<2-3 sentence overview>

## <Section 1>
<Structured content with all specifics preserved>

## Key Details
- **Decisions**: <every decision made, with rationale>
- **Open questions**: <anything unresolved>

## References
- `src/auth/middleware.ts` — auth middleware, token validation logic
- `src/auth/session.ts:42` — session expiry configuration

Preservation Rules

Never summarize away:

Exact file paths and line numbers
Error messages and error codes
Numeric values, measurements, benchmarks
Command outputs and their results
Decisions and the reasoning behind them
Function, class, variable, and module names
URLs and references

When condensing verbose material, compression must be reversible: keep the pointer (file path, URL, commit hash) next to every condensed claim so the detail is one Read away.

Key Constraints

## References is always the last section, each entry annotated with why it is referenced. This manifest powers staleness detection by /select and the bundled script (scripts/wisci.py).
The ## References section is rebuilt from scratch on every write — current state, not accumulated history.
When in doubt during merge, preserve both versions and flag with .