synthesize-project-context
Skill2 starsChanged 2 months agoMITView on GitHub
Synthesize upstream artifacts (requirements, source evidence, architecture/dependency/design findings, readiness) into candidate entries for PROJECT-CONTEXT.md — the S10 synthesis step that the project-context carrier format lacks. Use when multiple upstream analyses have completed and the unified project context must be written or updated (S10 synthesize-project-context slot), or when asked to "synthesize project context" / "update project context from findings" / "write the project context from these reports".
- Reads credentials
In this repository30 other files it gives its agents.
- agents-mdSkill
- agent-workflow-engineeringSkill
- analyze-architectureSkill
- architecture-auditSkill
- ask-mattSkill
- check-readinessSkill
- ci-cd-and-automationSkill
- codebase-contextSkill
- decision-memorySkill
- delegation-protocolSkill
- dependency-analysisSkill
- diagnosing-bugsSkill
- dispatching-parallel-agentsSkill
- documentation-maintenanceSkill
- engineering-closeoutSkill
- environment-discoverySkill
- github-pr-workflowSkill
- implementSkill
- interview-meSkill
- parallel-feature-developmentSkill
- planning-and-task-breakdownSkill
- project-contextSkill
- project-retrospectiveSkill
- release-managementSkill
- requesting-code-reviewSkill
- solution-architectureSkill
- subagent-driven-developmentSkill
- task-intakeSkill
- workflow-definitionSkill
- workflow-runnerSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 1 is distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 34
Read every listed artifact fully.
- Very specific: line 49
Read existing ## Active entries of PROJECT-CONTEXT.md.
- Very specific: line 64
Ask once: approve all / adjust (specify rows) / reject (specify rows).
- Very specific: line 68
Write approved entries into ## Active of PROJECT-CONTEXT.md (create the file with the standard header if absent).
- Distinctive: line 72
Before declaring done, all must hold:
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Synthesize Project Context
When to Use
- The S10 slot: after S03 (requirements), S05 (source evidence), S06 (architecture findings), S07 (dependency findings), S08 (design findings), S09 (readiness decision) outputs exist, before S11 planning.
- Any bulk update of
PROJECT-CONTEXT.mdfrom multiple findings (not single-entry adds — those go throughproject-context add).
Inputs
- Paths to upstream artifacts (provided at invocation): requirements / source-evidence / architecture-findings / dependency-findings / design-findings / readiness-decision. Any subset is valid.
- Existing
PROJECT-CONTEXT.mdat repo root, if present.
Outputs
- Updated
PROJECT-CONTEXT.md(approved candidates appended to## Active). - Synthesis summary: entries added / rejected (with reason) / merged with existing.
Workflow
Step 1 — Read upstream artifacts
Read every listed artifact fully. Missing files are marked [MISSING] in the summary — do not fabricate entries for them.
Step 2 — Extract candidate entries (inclusion gate)
For each finding in each artifact, apply the gate. PASS requires ALL four:
- Decision-relevant — affects future agent decisions (not a one-off fact of the current task)
- Durable — still likely relevant in 90 days (the default expiry)
- Non-duplicative — not self-evident from code, README, or AGENTS.md
- Safe — not a secret, credential, or .env content
Any failure → REJECT, and record the reason in the summary (transparency, not silence). When in doubt, reject: the cost of a stale entry is paid by every future session that reads it.
Step 3 — Deduplicate and boundary-check
- Read existing
## Activeentries ofPROJECT-CONTEXT.md; absorb candidates that restate an existing fact (merge, don't duplicate). - Boundary rules — content that belongs elsewhere is excluded here:
- Rules, commands, conventions →
AGENTS.md(S20) - Formal documentation →
docs/(S21) - Long-term decisions with rationale → decision records (e.g.
decision-memoryskill / ADRs) - This file holds informal but decision-relevant project knowledge: gotchas, half-finished state, historical constraints, warnings.
- Rules, commands, conventions →
- For each surviving candidate assign: source (which artifact), scope (
global| module/area name), expiry (default 90 days from today;neveronly for conventions that must not expire).
Step 4 — Batch confirm
Present ALL candidates at once as a table:
| # | Candidate entry | Source | Scope | Expiry |
Ask once: approve all / adjust (specify rows) / reject (specify rows). Do NOT run project-context add per entry for bulk synthesis — its per-entry confirmation flow does not scale (measured in trial F1).
Step 5 — Append
Write approved entries into ## Active of PROJECT-CONTEXT.md (create the file with the standard header if absent). Follow the exact entry format from the project-context skill.
Step 6 — Verify (completion gate)
Before declaring done, all must hold:
- Every entry has source, scope, and expiry
- No duplicates with existing Active entries
- No entry duplicates AGENTS.md / docs content
- Active count within the 50-entry / ~200-line cap (warn if approaching; suggest
project-context cleanup)
Anti-Patterns
- Extracting everything — the gate exists to prevent context rot; a bloated file gets ignored.
- Per-entry confirmation for bulk synthesis — batch confirm once, adjust by row.
- Writing rules/commands — belongs in AGENTS.md, not the context file.
- Fabricating entries for missing artifacts — mark
[MISSING]and move on. - Skipping expiry — default 90 days;
neverrequires explicit justification. - Duplicating an existing Active entry — merge or reject; duplicates halve the file's signal.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

